BIS Working PapersNo 577
Are Star Funds Really Shining? Cross-trading and Performance Shifting in Mutual Fund Families by Alexander Eisele, Tamara Nefedova, Gianpaolo Parise
Monetary and Economic Department
August 2016
JEL classification: G11, G14, G23
Keywords: mutual funds, cross-trading, performance shifting, conflict of interests
BIS Working Papers are written by members of the Monetary and Economic Department of the Bank for International Settlements, and from time to time by other economists, and are published by the Bank. The papers are on subjects of topical interest and are technical in character. The views expressed in them are those of their authors and not necessarily the views of the BIS.
This publication is available on the BIS website (www.bis.org).
© Bank for International Settlements 2016. All rights reserved. Brief excerpts may be reproduced or translated provided the source is stated.
ISSN 1020-0959 (print) ISSN 1682-7678 (online)
Are Star Funds Really Shining?Cross-trading and Performance Shifting in
Mutual Fund Families
Alexander Eisele, Tamara Nefedova, Gianpaolo Parise⇤
April 20, 2016
Abstract
The majority of financial trades take place in open and highly regulated mar-kets. As an alternative venue, large asset managers sometimes o↵set the tradesof a�liated funds in an internal market, without relying on external facilities orsupervision. In this paper, we employ institutional trade-level data to examinesuch cross-trades. We find that cross-trades used to display a spread of 46 basispoints with respect to open market trades before more restrictive regulation wasadopted. The introduction of tighter supervision decreased this spread by 59 ba-sis points, bringing the execution price of cross-trades below that of open markettrades. We additionally find that cross-trades presented larger deviations frombenchmark prices when the exchanged stocks were illiquid and highly volatile,during high financial uncertainty times, and when the asset manager had weakgovernance, large internal markets, and a strong incentive for reallocating perfor-mance. Finally, we provide evidence suggesting that cross-trades are more likelythan open-market trades to be executed exactly at the highest or lowest price ofthe day, consistent with the ex post setting of the price. Our results are consis-tent with theoretical models of internal capital markets in which the headquartersactively favors its “stars” at the expense of the least valuable units.
⇤Eisele is at the Swiss Finance Institute and the University of Lugano, Nefedova is at the Universite Paris-Dauphine, andParise is at the Bank for International Settlements. A part of this paper was written while Nefedova was at NYU Stern Schoolof Business and Parise was a visitor at Harvard Business School. We are thankful for suggestions and useful comments toGiovanni Barone-Adesi, Utpal Bhattacharya, John Campbell, Ben Cohen, Truong Duong, Inh Tran Dieu, Dietrich Domanski,Rudiger Fahlenbrach, Francesco Franzoni, Laurent Fresard, Rajna Gibson, Robin Greenwood, Harald Hau, Terrence Hender-shott, Augustin Landier, Eric Nowak, Kim Peijnenburg, Alberto Plazzi, Sofia Ramos, Christian Upper, David Schumacher,Suresh Sundaresan, Youchang Wu, and participants of seminars at Harvard Business School, the Bank for International Set-tlements, the Universite Paris-Dauphine, the EFA 2015 meeting in Vienna, the AFA 2014 meeting in Philadelphia, the AFFI2013 and 2015 meetings, the EFMA 2013 meeting in Reading, the Lugano Corporate Finance 2012 workshop, the Gerzenseedoctoral seminar, the Geneva conference on Liquidity and Arbitrage Trading 2012. We are also grateful to several practitionersfor valuable information on industry practices. We acknowledge financial support from Swiss National Science Foundation andthe Swiss Finance Institute. The views expressed in the paper are those of the authors and do not necessarily reflect those ofany institution with which they are a�liated. Contact at: [email protected]
1
According to recent estimates, around 40% of all U.S. stock trades takes today place outside
of public exchanges, up from 16% in 2008.1 Alternative trading practices mostly include dark
pool trading, internalizers, and cross-trades. Cross-trades – trades o↵set internally among
sibling funds a�liated to the same asset manager (or fund family) – are permitted under Rule
17a-7 of the U.S. Investment Company Act because they can in principle limit transaction
costs and commissions, benefiting the final investors. Anecdotal evidence warns however
that cross-trading might magnify the agency problem that arises when clients delegate the
investment decisions to the asset manager. For instance, a recent Financial Times article
reports a number of comments by industry participants on dubious cross-trading practices
including the following: “I’m aware that it happens, generally in equity funds but not always.
I suspect it’s quite widespread” and “It has happened many times in the past, often in times
of market pressure (...). In 2008 it was one way to ensure that prime money market funds
would be protected”.2
The task of investigating cross-transactions presents however an empirical challenge: most
institutional investors are obliged to disclose their holdings at a quarterly frequency only,
which makes it impossible to distinguish cross-trades from trades executed in opposite di-
rections but with external counterparties. In this paper, we use a sample of trades executed
by American asset managers from 1999 to 2010 in order to be able to identify cross-trades
correctly. Specifically, in our empirical analysis we look for pairs of trades originated from
the same asset manager, in the same stock, same quantity, executed exactly at the same day,
time, and price, but displaying opposite trading directions. We furthermore use data on com-
missions to verify the quality of our identification procedure, as cross-trades are supposedly
significantly cheaper than open market trades.
The exact identification of cross-trades allows us to address three so far unanswered ques-
tions. First, do cross-trades really minimize the transaction costs borne by the final investor?
Second, does the pricing of the cross-trades vary with market, manager or stock characteris-
1“Dark markets may be more harmful than high-frequency trading”Reuters – April 7, 2014.2“No Surprise at Backroom Dealing Charge” Financial Times - December 16, 2012.
2
tics? Third, how does cross-trading a↵ect the di↵erence in performance between “star” and
“junk” funds (i.e., funds of relatively high/low importance from a family perspective)?
We conduct three sets of empirical tests to address these questions. First, we explore
how cross-trades were priced. The rationale of allowing cross-trades to benefit investors
suggests that the spread between the execution price and the market price of the stock at
the moment of the transaction (hereafter the “Execution Short f all”) is low for cross-trades
because transaction costs are minimized. Conversely, we find that cross-trades used to exhibit
an execution shortfall 18 basis points higher than for trades executed in the open market after
controlling for the size of the trade and stock, time, and family fixed e↵ects. This extra cost
involved in the cross-transaction reallocated performance between the two parties involved
in the trade (e.g., one fund buys at a discount from one of its siblings.) This cost however
disappeared when restrictive regulation was introduced.
Second, we explore how di↵erent stock characteristics and market conditions a↵ect cross-
trades. We find that cross-trades in illiquid and highly volatile stocks presented more signif-
icant deviations from benchmark prices. Additionally, we provide evidence suggesting that
the execution price of cross-trades may sometimes have been set ex post to the highest or
lowest price of the day (which we refer to in this paper as “backdating”).3 Furthermore, we
investigate how the execution shortfall correlated with fund family characteristics. Our null
hypothesis is that family characteristics were irrelevant in explaining how cross-trades get
priced. Alternatively, if cross-trades were used to shift performance in an opportunistic way,
we should find a higher execution shortfall within families for which agency problems were
more relevant – namely, families in which governance is weak and family incentives diverge
from investors’ interests (Massa (2003), Nanda, Wang, and Zheng (2004), Chuprinin, Massa,
and Schumacher (2015)). Exploring the cross-section of cross-trades, we find that the exe-
3Lower regulatory scrutiny on cross-trading activity compared to open market trades could more easilyallow institutions to arbitrarily set ex post the execution price of the cross-trade at the price of the dayat which the greatest performance would have been reallocated among trading counterparties. Consistentwith this argument, we show that cross-trades were significantly more likely than open market trades to beexecuted exactly at the highest/lowest price of the day.
3
cution shortfall used to be significantly higher for cross-trades executed in families in which
governance was weak, there was a high number of siblings and some funds were significantly
more expensive than others.4
Finally, we explore how cross-trading activity a↵ected the di↵erence in performance be-
tween star and junk funds. Building on previous theoretical work on internal capital markets
we can formulate two hypotheses. On the one hand, mutual fund complexes may work as
conglomerates in which strong divisions end up subsidizing weak ones (Stein and Scharf-
stein (2000)). In this context, powerful managers of poorly performing funds may force star
funds to engage in ine�cient cross-subsidization via badly priced cross-trades. The resulting
outcome would be performance smoothing across di↵erent funds within the same fund family.
On the other hand, the corporate headquarters of a multi-division company has control
rights that enable it to engage in “winner-picking,” i.e., to actively shift resources to few
successful projects (Stein (1997)). Similarly, fund families may use cross-trades to allocate
extra performance to a number of popular or expensive funds. A large body of research
on mutual funds suggests that outperformers, while attracting disproportionate inflows to
themselves,5 also have positive spillover e↵ects on the other siblings in the family (Nanda,
Wang, and Zheng (2004), Brown and Wu (2015)). This would make it potentially optimal
from a family perspective to penalize less important funds to pump up the returns of their
star funds. Consistent with Gaspar, Massa, and Matos (2006), we show that star funds’
performance used to benefit from the extent of cross-trading activity in the fund family at
the expense of junk funds.
However, both reverse causality and omitted variables may a↵ect the validity of our re-
sults. To tackle a number of identification concerns, we use an exogenous change in fund
families’ internal governance to assess the impact of potential cross-trading activity on ex-
4A high number of a�liated funds creates incentives for tournament behavior (Brown, Harlow, and Starks(1996), Kempf and Ruenzi (2008)) and allows a fund family to transfer performance via cross-trades in alarge internal market, while families with high heterogeneity in funds’ importance are those with the strongestincentives to reallocate performance (see Gaspar, Massa, and Matos (2006)).
5There is abundant evidence that outperformers attract greater inflows, e.g., Chevalier and Ellison (1997),Sirri and Tufano (1998), Agarwal, Gay, and Ling (2014).
4
ecution shortfall. In 2004, new regulation enforcing more e↵ective compliance policies was
introduced after an investigation uncovered widespread malpractice in industry practices. We
therefore compare how the pricing of cross-trades responded to increased regulatory scrutiny
with respect to open market trades (as open market trades should be una↵ected by such
a change in regulation), thereby providing evidence in favour of a causal interpretation of
our results. Additionally, we show that cross-trading went from peaks above 6% of the total
trading activity before the new regulation was introduced to below 1% afterwards, suggesting
that greater supervision may significantly a↵ect the incentive to rely on cross-trading activity.
Furthermore, the execution shortfall fell below that of open market trades.
In this paper, we make two main contributions to the existing literature. First, this is
to the best of our knowledge the first paper providing direct evidence on the pricing and
characteristics of actual cross-trades. The use of cross-trades is pervasive in the mutual
fund industry, and the regulator has decided to allow exemptions for cross-trading in other
industries as well.6 Therefore, a study on cross-trading activity provides important policy
implications, while improving our understanding of incentives at the fund family level. Our
paper is the first to show that cross-trades in the past seem to have been significantly mis-
priced and potentially backdated. Tying cross-trade level data to fund performance, we find
that cross-trading potentially boosted the risk-adjusted performance of star funds by around
1.7% per year on average (causing an equivalent loss for the least important funds). This
result casts doubt on the fraction of performance delivered by mutual funds that is really due
to skill.7
Second, we show that the introduction of tighter supervision in 2004 resulted in a sig-
nificant decrease in both the frequency of cross-trades and the average execution shortfall.
6See, e.g., the cross-trading exemptions under section 408(b)(19) added to ERISA on August 17, 2006 bythe U.S. Department of Labor.
7In general a large body of literature has been devoted to the study of mutual fund performance see,e.g., Kacperczyk, Sialm, and Zheng (2005), Kacperczyk, Sialm, and Zheng (2008), Massa and Patgiri (2009),Kempf, Ruenzi, and Thiele (2009), Huang, Sialm, and Zhang (2011), Ferreira, Keswani, Miguel, and Ramos(2012), Chen, Hong, Jiang, and Kubik (2013), and Brown and Wu (2015). Our results suggest that cross-trading activity significantly contributes to explaining the historical cross-section of fund returns.
5
A lower deviation from benchmark prices limits, but does not necessarily exclude, potential
performance redistribution. However, careful regulatory scrutiny seems to be highly e↵ective
in limiting both the extent of mispricing and the incentive to cross-trade. The focus of our
analysis is on cross-trading activity. Cross-trades are however only one alternative to trading
in open exchanges. For instance, large asset managers increasingly rely on “dark pools” and
other opaque trading venues.8 Overall, our findings point to potential risks posed by the
increasing popularity of unsupervised and less regulated trading.
This paper proceeds as follows: Section I briefly reviews the related literature and high-
lights the contributions of our paper. Section II provides information on the data and de-
scribes how cross-trades are identified. Section III explores how cross-trades are priced, o↵ers
evidence from the cross-section of cross-trades, and tests the backdating hypothesis. Section
IV documents the impact of cross-trading activity on fund performance. Section V provides
further results and robustness checks. Section VI concludes.
I. Related Literature
Previous literature hypothesizes the presence of cross-subsidization in the money management
industry. Gaspar, Massa, and Matos (2006) find that when sibling funds trade in the opposite
direction, the performance of high-value funds (expensive or successful funds) is boosted
and the performance of low value funds decreases. The authors claim that this pattern is
consistent with performance shifting via cross-trading. Conversely, Schmidt and Goncalves-
Pinto (2013) argue that fund families might systematically shift performance via cross-trades
from popular funds (funds attracting positive investor flows) to distressed funds by absorbing
flow-induced fire-sales. In both cases, the authors focus on opposite trades computed from
quarterly snapshots of mandatory fund filings.9 In this regard, Gaspar, Massa, and Matos
8A growing academic literature explores the implications of alternative trading venues (e.g., Zhu (2014)and Comerton-Forde and Putnins (2015)), yet data availability is usually a major issue.
9E.g., if fund A buys 1000 of a stock in January and fund B belonging to the same fund family sells800 of the same stock in March, the two funds are assumed to be cross-trading for 800 by related studies,significantly overestimating the number of cross-trades.
6
(2006) state the following: “we should make clear from the start that we can only provide
evidence that is limited by the level of information disclosure to which mutual fund activities
are subject”.10
Additionally, a large literature explores incentives at the fund family level aside from the
cross-trading setting. Bhattacharya, Lee, and Pool (2013) show that a�liated funds of funds,
i.e., funds that can only invest in other funds in the family, overweight their holdings in funds
that are forced to sell.11 Evans (2010) finds that funds outperform while “incubated” but
such outperformance disappears after the funds are open to investors; Chuprinin, Massa, and
Schumacher (2015) find that in-house funds outperform outsourced funds by 0.85% annually
consistent with the hypothesis of preferential treatment; and Nanda, Wang, and Zheng (2004)
shows that fund families have a high incentive to start several new funds, increasing their
chance of producing “star funds” (i.e., funds that outperform by chance).
While the evidence provided in previous studies is suggestive of opportunistic performance
shifting via cross-trades, it does not necessary rule out three sets of alternative explanations.
First, di↵erential skill or resources might explain why star funds are on average more likely
to trade in the “right” direction (Guedj and Papastaikoudi (2008)). Second, reverse causality
might be an issue as di↵erence in performance may lead a�liated funds to trade in opposite
directions. Third, other within-quarter unobserved actions (i.e., actions that cannot be di-
rectly inferred by quarterly filings), such as security lending, timing of interim trades, IPO
allocations, and window-dressing behavior, may contribute to explaining the gap in perfor-
mance between star and junk funds (Kacperczyk, Sialm, and Zheng (2008)).
Additionally, previous research cannot identify cross-trades and therefore leaves necessar-
ily unanswered questions related to how cross-trades are priced and how their pricing relates
10Other papers looking at cross-trading activity are Chaudhuri, Ivkovich, and Trzcinka (2012) and Casavec-chia and Tiwari (2015), however also those papers do not have information on pricing, timing, exact volumeof the transactions and stock characteristics.
11While this result may seem apparently in contrast with our finding, the settings of the two papers are verydi↵erent. We consider equity funds and not funds of funds, we focus on asset trades, while Bhattacharya,Lee, and Pool (2013) explore investments in the shares of distressed funds. We focus on all mutual fundfamilies, while Bhattacharya, Lee, and Pool (2013) consider only the subset of mutual fund families thatincludes a�liated funds of funds.
7
to stock, market, and family characteristics. Using trade-level data, we show that i) cross-
trades might have been opportunistically priced and occasionally backdated, ii) cross-trades’
pricing and frequency changed drastically when regulatory scrutiny increased, and iii) devi-
ations from benchmark prices seemed more relevant when the assets exchanged were illiquid
and the cross-trades were executed in high uncertainty times. Anand, Irvine, Puckett, and
Venkataraman (2012) show that trading costs have a significant e↵ect on performance and on
the persistence of relative fund performance, in this paper we find that cross-trades present
substantial di↵erences in trading costs from open market trades.
Finally, tying our trade-level data to fund-level returns we provide evidence from ac-
tual trades for the favoritism hypothesis proposed in Gaspar, Massa, and Matos (2006) and
Chuprinin, Massa, and Schumacher (2015) (instead of indirect evidence from quarterly hold-
ings and returns). Conversely, our results cast doubt on the hypotheses that star funds pro-
vide insurance to distressed funds on average (see, e.g., Schmidt and Goncalves-Pinto (2013))
and on the assumption that funds mostly cross-trade illiquid positions to lower transaction
costs (Goncalves-Pinto and Sotes-Paladino (2015)).
II. Data, Identification of Cross-trades, and
Summary Statistics
In our analysis, we focus on mutual fund families as a laboratory. Cross-trades are common
also in other industries, however the mutual fund setting allows us to obtain data from a
large number of sources. This section describes the di↵erent datasets we use for our analysis.
A. Trade-Level Data
We obtain trade-level data from Abel Noser Solutions/ANcerno, a consulting firm that works
with institutional investors monitoring their trading costs. Batches of data sent by its clients
8
include all executed trades for the whole period covered by the batch.12 Previous research
has shown that ANcerno institutional clients constitute approximately 8% of the total CRSP
daily dollar volume (Anand, Irvine, Puckett, and Venkataraman (2012)) and that there is no
survivorship or backfill bias in the data (see, e.g, Puckett and Yan (2011)). Despite ANcerno
claims that all trades are disclosed, we cannot rule out that clients opportunistically choose
which trades to submit or that they intentionally misreport execution prices. However,
strategic reporting would bias the results against finding evidence for opportunistic pricing
of the cross-trades. Therefore, we conclude that our results are unlikely to be a↵ected by
opportunistic reporting or, in any case, under-represent the real extent of mispricing.
ANcerno provides us several variables useful for our investigation: stock identifier (cusip),
trade date, execution price, execution time,13 number of shares executed, side of the trade
(i.e., buy or sell), price of the stock at the time of the execution, commissions paid, and
volume-weighted average price of the day (VWAP). ANcerno provides information both at
the order (called ticket in Anand, Irvine, Puckett, and Venkataraman (2012)) and at the
trade level. In particular, each order can be broken down in a number of trades executed
at di↵erent times of the day (or in some extreme cases across di↵erent days). We find the
number of trades to be more than double the number of orders in our sample in line with
Anand, Irvine, Puckett, and Venkataraman (2012). As the relevant benchmark for cross-
trades is the price at which the trade is executed (while the price at the placement of the
order is irrelevant), we conduct most of our analysis at the trade level.
This information is sent to ANcerno by its di↵erent clients.14 The identity of the clients
is always anonymized. Importantly, while the client is anonymized, the asset manager is not.
For a limited period of time in 2010–early 2011 ANcerno provided its academic subscribers
with the identification table “MasterManagerXref” including unique codes (managercodes)
12Examples of other empirical studies using ANcerno include Chemmanur, He, and Hu (2009), Anand,Irvine, Puckett, and Venkataraman (2012).
13The time variable is based on a 24-hour day and is precise to the minute (i.e., not the second) level.14A client can either be a single fund or a fund manager managing multiple funds or, alternatively, a money
manager which is managing a portfolio on behalf of the client.
9
with associated names of the asset manager to whom they were a�liated. It is important to
mention that the set of provided identification files is subscription-specific. The sample used
in this study is constructed using the fullest set of identification files provided by ANcerno,
to which earlier and later ANcerno subscribers do not have access to. The full file includes
1,088 asset managers. Additional identification files “ManagerXref” and “BrokerXref” have
the necessary variables to link managing companies and brokers to the trades. The same
identification files allow us to match ANcerno data with the Thomson Reuters database
unambiguously. In particular, we hand-match fund families from ANcerno to 13F/S12 by
name. For instance, the match table provided by ANcerno includes a manager name, e.g.,
“XYZ Capital”15 and a managercode, e.g., 10. This allows us to match the managercode
number to a number of trades in ANcerno executed by funds a�liated to XYZ Capital.
Our matched database spans the time interval from 1999 to 2010 as ANcerno stopped to
provide any identification of the asset manager after that date. Hence, we cannot conduct
our analysis in the post-2010 period. Unfortunately, while ANcerno provided us with unique
asset manager (family) identifiers, it does not make available unique fund identifiers. There-
fore, we are able to link with certainty each cross-trade to the asset manager to which the
fund is a�liated but not to the fund itself. Our contact in ANcerno explicitly excluded that
it is possible to identify with certainty specific funds using ANcerno data. Some fund/fund
manager names are occasionally reported by ANcerno but they appear to be highly unreliable
or incomprehensible. We are therefore able to establish that two trades from asset manager
XYZ Capital are internally crossed with each other but we can only provide suggestive evi-
dence on the exact identity of the two funds that are cross-trading. However, as our analysis
is mostly conducted at the trade level, the exact identity of the funds is irrelevant as long as
we are able to ensure that two trades are o↵set within the same fund family.
15XYZ Capital is not actually a real asset manager included in our sample.
10
B. Identifying Cross-trades
A cross-trade is a transaction in which a buy and a sell order for the same stock coming
from the same fund family is conducted without going through the open market. We identify
cross-trades in our database as transactions occurring i) within the same fund family, ii) in
the same stock, iii) at the same time of the same day, iv) at the same price, and having v)
the same volume of the trade but in opposite trading directions. For instance, a buy trade
of 1,000 Apple stocks executed on January 2nd, 2010, at 10:05 a.m. for $101 is classified as
a cross-trade only if we have in our sample a sell trade of 1,000 Apple stocks coming from
the same fund family and executed on January 2nd, 2010, at 10:05 a.m. for $101. It is in
theory possible, but highly unlikely, that two funds belonging to the same fund family make
exactly the same trade in opposite directions, at the same time, by chance. Mutual funds
do not trade at very high frequencies and usually a�liated funds rely on the same research
which leads them to rarely trade in opposite directions (Elton, Gruber, and Green (2007)).
To check the reliability of our matching procedure, we furthermore compare commission
costs of open market trades with commission costs of the trades we identify as cross-trades.
In particular, commissions for cross-trades should be zero or extremely small (the broker
does not need to find a counterparty for the trade, although sometimes a commission is due
for bookkeeping services). We find that the average commission ($/share) for cross-trades
in our sample is 0.0016 (the median commission is 0), while it is around 0.0245 for open
market trades (see Table I, Panel A). The number reported by Anand, Irvine, Puckett, and
Venkataraman (2012) is slightly higher (0.028). We however find basically the same number if
we limit our sample to the same time interval (1999-2008). If we compare dollar commissions
per dollar trade, the average value is 11 basis points for open market trades and 1 basis
point for cross-trades. The di↵erence is statistically significant at the 1% level. In particular,
commissions are 0 for more than 90% of the trades we classify as cross-trades, suggesting
that our algorithm identifies cross-trades with high precision.16
16Results are analogous when considering as cross-trades only transactions where no commissions have
11
This identification procedure overcomes the main limitation of proxies computed from
quarterly or semi-annual snapshots employed by Gaspar, Massa, and Matos (2006), Schmidt
and Goncalves-Pinto (2013), and Chuprinin, Massa, and Schumacher (2015). Using our
approach, opposite trades recorded in the same quarter but occurring in di↵erent days/times
and having di↵erent volumes are not considered as cross-trades. Therefore, in our trade-level
analysis we compare the pricing of cross-trades (CT Dummy equal to one) versus open market
trades (CT Dummy equal to zero). In the latter part of the paper, we provide some suggestive
evidence on the impact of cross-trading on fund performance. When we run regressions at
the fund level, our main explanatory variable is CT % f ,t : the monthly total dollar volume of
cross-trades executed by family f in month t as a proportion of the total dollar volume of
trades (open-market trades plus cross-trades) executed by family f in month t.
C. Execution Shortfall
ANcerno provides us with a series of benchmark prices for each trade in our sample. Rule
17a-7 of the U.S. Investment Company Act establishes that cross-transactions should occur
at the “current market price” of the security. We then focus on the market price at the
moment of the execution as the main benchmark as this seems to be the closest to what
Rule 17a-7 prescribes.17 In using the price at execution as a benchmark, we rely on the
information provided by ANcerno. However, in a limited number of cases ANcerno arbitrarily
sets the execution time of the trade at the end/beginning of the day if the time is missing
in the information provided by the institutional investor (see Anand, Irvine, Puckett, and
Venkataraman (2012)). If the execution time reported is incorrect, this could potentially add
significant noise to our results. Assuming that misreporting is random there is no reason why
been paid. Occasionally, commissions are not charged also for normal trades. Therefore, the reporting ofzero commissions is neither a necessary nor su�cient condition for a trade to be considered a cross-trade.
17The time of the execution is provided at the minute level. However, trades can be executed at di↵erentseconds of the same minute. This would create by construction a spread between the execution price of atrade and its benchmark. Since this should a↵ect in the same way open market trades and cross-trades (theexact execution time within the minute should be random for both), it does not compromise the validity ofour results.
12
cross-trades should be systematically set at the highest price. Conversely, if misreporting is
strategic this should limit the incidence of mispricing cases and bias our results downward.
Furthermore, to make sure this does not significantly a↵ect our results we replicate our
analysis dropping trades executed exactly at the opening or closing price of the day, finding
analogous results. Anyway, our ANcerno contact assured us that this problem only a↵ects
an extremely limited number of trades, mostly reported by pension funds.
Cross-trades should minimize the impact of trading costs and commissions on the ex-
ecution price, limiting deviations from the price quoted on the market (which is our main
benchmark). Therefore, we define Execution Short f all as the absolute value of the deviation
from the benchmark price scaled by the benchmark price itself. Consideration of the abso-
lute value of the deviation from the benchmark is necessary in our setting. In fact, for each
cross-trade our sample includes two twin trades with opposite execution shortfalls that would
cancel each other out if signed values were considered. Hence, we define Execution Short f all
as:
Execution Short f all j,i,t =|Pj,i,t �Pi,t |
Pi,t, (1)
where Pj,i,t is the execution price of trade j, in stock i, at execution time t; while Pi,t is the
price of stock i in the market at time t. Results using alternative benchmarks are presented
in Section V.18
18To verify the robustness of our results and rule out the possibility that they are driven by misreportingof the execution time, we replicate our analysis using the volume-weighted average price of the day (VWAP)and the open price of the day as alternative benchmarks (this does not require us to use the executiontime variable at all). Some studies argue that the price at execution should be compared with the VWAP,which is also the most popular benchmark among practitioners (see Berkowitz, Logue, and Noser (1988),Hu (2009), and Anand, Irvine, Puckett, and Venkataraman (2012)). However, other studies warn aboutpotential shortcomings in the use of VWAP as a benchmark (see, e.g., Madhavan (2002) and Hasbrouck(2007)). For instance, large trades are more likely to be executed exactly at the VWAP. Therefore, followingBusse, Chordia, Jiang, and Tang (2015) we replicate our analysis also using the open price of the day as abenchmark (see Section V). In all cases results are qualitatively similar.
13
D. Fund-level Data
The main focus of our analysis is on trades. However, the exact identification of cross-trades
allows us to provide some evidence at the fund level by linking our sample to CRSP mutual
fund data via the asset manager identity. We therefore obtain measures of a mutual fund’s
size, its fees, and its flows. Following Gaspar, Massa, and Matos (2006) we compute fees
as 1/7( f rontload+rearload)+expense ratio. We compute fund flows following the literature
(see, e.g., Coval and Sta↵ord (2007)):
Flowi,t =T NAi,t � (1+ reti,t)T NAi,t�1
T NAi,t�1, (2)
where T NAi,t are the total net assets under management and reti,t is the monthly return
of fund i in month t. At the family level, we compute the family size, the intra-family return
dispersion, the intra-family expense ratio dispersion, and the intra-family dispersion in fees.
Family Size is defined as the (log of the) sum of the individual funds’ assets at the beginning
of the month. We compute intra-family Return Dispersion as the standard deviation of the
returns of a�liated funds in the previous month, Expense Ratio Dispersion as the standard
deviation of the expense ratios, and Fees Gap as the di↵erence between the highest and
lowest fee charged by funds a�liated to the asset manager in a given month.
Additionally, we compute the variable Siblings as the natural log of the number of eq-
uity funds belonging to the same family f in month t (Pollet and Wilson (2008)). We use
Thompson Reuters investment objective codes to identify the investment style for each fund.
Following Ferris and Yan (2009), we also build a proxy of governance based on precedent
infringements. In particular, we argue that fund families investigated by the SEC for ille-
gal practices potentially harming investors (besides cross-trading related practices) are more
likely to have weak governance. Consistent with this argument, Dimmock and Gerken (2012)
show that past legal violations have significant power to predict future fraud. Therefore,
we manually examine SEC administrative proceedings and the Wall Street Journal Mutual
14
Funds Scandal Scorecard to categorize each fund family as having either Weak or Strong
Governance.19
E. Summary Statistics
Sample statistics on the matched fund sample are reported in the Appendix (see Table A.I).
Columns 2 to 4 show statistics from the CRSP mutual fund-Thomson Reuters match. The
intersection between the two samples leaves us with 2,351 funds, organized into 452 fund
families. The average mutual fund size is USD 1,258 million, while the average mutual fund
family size is USD 39,531 million. The average fund family includes 17 equity funds.
Matching our sample of mutual funds to the ANcerno database decreases our sample
size significantly. The final number of asset managers in our sample is 203 fund families
managing 1,393 mutual funds. In particular, our matched sample contains 45% of the mutual
fund families in the CRSP-TR dataset. However, such families account for 59% of the funds.
Our sample is biased toward large institutions because the smallest families are less likely
to rely on ANcerno’s services (this bias has been recognized also by previous studies (see,
e.g, Puckett and Yan (2011)). Additionally, the funds in our final database perform slightly
better than funds in the CRSP database.20 This di↵erence may be explained by the fact that
funds belonging to large fund families perform better on average (Chen, Hong, Huan, and
Kubik (2004)).
To limit the sample size in our empirical analysis on trade-level data, we extract three
random samples consisting of 1% of the original ANcerno sample and retain only observations
for which we have all control variables21, we produce results for all 3 samples and report
results from sample 2 since these are the weakest.22 Therefore, results in the paper are likely
19We focus on investigations instead of final court rulings because more than 90% of the investigations endup in out-of-court settlements.
20Average flow of 0.28% in the matched sample versus 0.09% in CRSP; average monthly raw return of0.42% versus 0.37%; and average monthly alpha of 0.03% versus 0.00% (see Table A.I in the Appendix.)
21This procedure is not uncommon in the asset pricing literature (see, e.g., Ben-David and Hirshleifer(2012)).
22Results from samples 1 and 3 are anyway very similar and are reported in the Appendix (see Table A.II).
15
to provide a lower bound for the mispricing of cross-trades. Our sample consists of 966,186
trades out of which we classify 7,368 as cross-trades and 958,818 as open market trades.
Table I, Panel A reports the number of observations (Column 1), and average values for all
main variables in the full sample (Column 2), average values keeping open market trades
only (Column 3), cross-trades only (Column 4), the di↵erence between open market trades
and cross-trades (Column 5), and t-statistics for the null hypothesis of equality in the means
of open market trades and cross-trades (Column 6). The statistics show that cross-trades
are significantly bigger than normal trades both in share and dollar volume. Additionally, in
general cross-trades involve stocks that present higher bid-ask spreads, are more volatile, and
are bigger: they exhibit on average higher market capitalization and are more likely to be
included in the S&P500 index. The fact that most of the cross-trades occurs in large market
capitalization stocks is probably due to the high overlap of large stocks in funds’ portfolios.
Cross-trades are significantly cheaper than regular trades in terms of commissions (more
than 90% of them display 0 commission) reassuring on the quality of our identification proce-
dure. However, cross-trades exhibit significantly higher execution shortfall than open market
trades (0.84% versus 0.64%). This may be due to the fact that cross-trades are used to
strategically shift performance between counterparties or simply to the di↵erence in stock
characteristics. Multivariate analysis is employed in the next section to disentangle between
these two hypotheses. Our numbers are higher than those reported in Anand, Irvine, Puck-
ett, and Venkataraman (2012). This is due to three main di↵erences in how we compute
our shortfall measure that arise naturally from our di↵erent research design. First, we al-
ways compute the absolute value of Execution Short f all while Anand, Irvine, Puckett, and
Venkataraman (2012) do not (we focus on the deviation from the benchmark irrespectively of
the direction of the trade since each cross-trade is a zero-sum game in which there is a winner
and a loser party, therefore for our research design it would not make sense to compute the
signed deviation). Second, we use the market price at the execution instead of the price
at placement as a benchmark (Rule 17a-7 of the U.S. Investment Company Act states that
16
cross-trades should be executed at the prevalent price at the moment of the trade). Third,
we look at trades and not at orders (a single order can be broken down in several trades
executed at di↵erent times, the relevant benchmark for our analysis will di↵er depending on
when each single trade is executed.)
In Panel B we report pairwise correlations among our main variables to make sure that
stock and trades characteristics have the impact of execution shortfall predicted by the theory.
Consistent with previous research, we find that Execution Short f all is positively correlated
with proxies of stock illiquidity and negatively correlated with proxies of size.
III. The Pricing of Cross-Trades
A. Cross-Trades and Execution Shortfall
Our empirical strategy uses cross-sectional variation to explore how cross-trades are priced
compared to trades executed in the open market. Rule 17a-7 of the U.S. Investment Com-
pany Act allows cross-trades subject to conditions of fair valuation of assets (“independent
current market price,” usually last sale market price) and fair treatment of both parties. The
Securities and Exchange Commission specifies that the adviser has a duty to, among other
things, “carefully consider” its responsibilities of best execution and loyalty to each fund.
In particular, a cross-trade should never occur when one party could obtain a better price
by going to the open market. Our null hypothesis is, therefore, that cross-trades exhibit a
significantly smaller execution shortfall than ordinary trades, as a higher deviation from the
benchmark price would suggest that one trading counterparty gets unfairly penalized. In our
analysis, we compare the execution shortfall of cross-trades with the execution shortfall of
open market trades controlling for trade/stock/time/family di↵erences.
Therefore, we run trade-level ordinary least square regressions of Execution Short f all
on CT Dummy, a dummy variable that takes a value of one when a trade is a cross-trade
and takes a value of zero when a trade is executed in the open market. The identification
17
strategy to pin down cross-trades is extensively explained in Section II.B. The choice of
the benchmark and the potential shortcomings of our approach are carefully discussed in
Section II.C. We additionally include stock, time, and fund family fixed e↵ects to absorb
time invariant di↵erences and we cluster errors at the time level to account for cross-sectional
heterogeneity. Formally,
Execution Short f alli, f ,t = b(CT Dummyi, f ,t)+G0Xi,t + gi + g f + gt + ei, f ,t , (3)
where i indexes the stock, t the time, and f the fund family. Xi,t is a vector of time-varying
stock-level controls, gt, gi, and g f are time,23 stock, and family fixed e↵ects, respectively. The
identification of an e↵ect forCT Dummyi, f ,t on Execution Short f all comes from the comparison
of cross-trades with otherwise similar trades that are not crossed.
Table II, Column 1 shows that the Execution Short f all in our sample is 19 basis points
higher for cross-trades compared to open market trades. This result is significant at the 1%
level (t-statistic of 5.44). A potential explanation for this di↵erence, however, is that cross-
trades are on average larger in volume. To be certain that our result is not driven by trading
volume, we include the volume of the trade24 as a control variable in specification (2). Table
II, Column 2 shows that a higher trading volume indeed a↵ects Execution Short f all, as the
coe�cient of our CT dummy decreases from 19 to 18 basis points, while still being significant
at the 1% level.25
However, time-varying stock characteristics may also have an e↵ect on Execution Short f all.
For instance, highly volatile and illiquid stocks usually display higher execution shortfall.
Therefore, we include stock level time-varying controls for di↵erent proxies of stock illiquid-
ity: the Amihud Ratio (Illiquidity), the ratio of one over the open price of the day (1/Price),23We use month-level fixed e↵ects to limit the number of dummies in our model. To include day fixed
e↵ects and cluster errors at the day-level actually yields economically and statistically stronger results.24We use the share volume of the trade instead of the dollar volume to avoid mechanical correlation with
the dependent variable as the price of the stock would be included both in the dependent and independentvariable. However, results using the dollar volume are similar.
25In a previous version of the paper, we reported lower estimates. The di↵erence is due to the inclusion offamily and stock fixed e↵ects in this version of the paper.
18
and the Bid-Ask spread (Bid-Ask Spread). Additionally, we include proxies for stock capital-
ization because bigger stocks display in general lower Execution Short f all. In particular, we
add a dummy for the inclusion in the S&P500 index (S&P500 Dummy) and the stock market
capitalization decile (Market Equity Decile). Finally, we control for the standard deviation
of stock daily returns (Volatility). However, the impact on the magnitude and significance
of the main coe�cient of interest is marginal at best. All signs of the control variables are
consistent with related research.26 Results obtained from the full specification model includ-
ing all control variables, indicate that cross-trades have a 18 basis points higher Execution
Short f all than open market trades.27 This result is significant at the 1% level (t-statistic of
5.37).
Our result is economically significant. The average percentage bid-ask spread (bid-ask
spread over bid) is 4 basis points in our sample. The marginal e↵ect of cross-trades on
the execution shortfall is 4.5 times higher. Conservative back-of-the-envelope calculations
suggest that cross-trading might have shifted performance for $1.8 million per day in the
mutual fund industry only.28 However, the exact impact of mispricing on fund performance
depends on additional factors, such as, the extent of cross-trading activity and the size of the
fund itself. Section IV provides suggestive estimates of the impact of mispriced cross-trades
on fund performance. Overall, this section provides evidence that adds to the results from
Kacperczyk, Sialm, and Zheng (2008) and Anand, Irvine, Puckett, and Venkataraman (2012)
in suggesting that unobserved actions occurring within a quarter, and therefore not captured
by obligatory fund filings, might have significant implications for performance.
26With the only exemption of Bid-Ask Spread which turns insignificant when the other proxies of stockilliquidity are included due to the high correlation among them.
27We obtain similar results using di↵erent fixed e↵ects, see Appendix.28This number is obtained by multiplying $169 billion (average daily trading dollar volume on the NYSE)
times 0.30 (roughly the total US equity held by mutual funds according to Investment Company 2015 factbook)times 0.02 (average cross-trading activity out of total dollar trading volume of mutual funds in our sample)times 0.0018 (marginal e↵ect of CT Dummy on execution shortfall). This number is likely to be a lower boundas considering only the NYSE significantly under-represents the total amount of trading activity.
19
B. Reverse Causality and Endogeneity
B.1. The Natural Experiment
One concern with our previous results is the direction of causality. A reverse causality argu-
ment suggests that instead of cross-trades explaining execution shortfall, it was high expected
execution shortfall that drove the decision of fund managers to cross-trade. Additionally,
omitted variables may a↵ect both Execution Short f all and the choice of a fund manager to
cross-trade. We address these concerns by using an exogenous increase of regulatory scrutiny.
On September 3, 2003 the New York State Attorney General Eliot Spitzer announced
the issuance of a complaint claiming that several mutual fund firms had arrangements allow-
ing trades that violated terms in their funds’ prospectuses, fiduciary duties, and securities
laws (the investigation led to what is commonly referred to as the “late trading scandal”).
Subsequent investigations showed that at least twenty mutual fund management companies,
including some of the industry’s largest firms, had struck deals permitting improper trading
(Zitzewitz (2006), McCabe (2009)). Importantly, most of the violations involved late-trading,
while none of the funds under scrutiny were charged with improper cross-trading.29
As a consequence of the scandal, in 2004 new rules were introduced and adopted by
the Securities and Exchange Commission (SEC) requiring fund families to implement more
stringent compliance policies. In particular, Rule 38a-1 under the Investment Company Act of
1940 forced investment companies to adopt and implement policies and procedures reasonably
designed to prevent violations of federal security laws and designate a chief compliance o�cer
responsible for administering such policies and procedures reporting directly to the board of
directors. Rule 206(4)-7 under the Investment Advisers Act of 1940 imposed equivalent
requirements on each adviser registered with the Commission. We contacted a number of
compliance o�cers at leading management companies to obtain more information about
the actual implications of the new regulations: they pointed out that the supervision of
29The late trading scandal has been used as a source of exogenous variation in other papers (see, e.g.,Anton and Polk (2014)). However, in this paper we are not interested in the late trading scandal per se, butmainly into the regulatory framework that was implemented as a response to the scandal.
20
cross-trading activity and the monitoring of cross-trades pricing became one of their key
responsibilities in 2004.
We argue that both the increased attention to improper trading practices in the industry
induced by the late trading scandal and the tightening of regulation led to a reduction in
potentially opportunistic cross-trading activity. This exogenous shock allows us, first, to
improve our estimation of the impact of potential cross-trading on performance. Second, it
permits us to estimate what proportion of cross-trades was executed primarily for potentially
opportunistic reasons.
The new rules became e↵ective on February 5, 2004 while the date companies were re-
quired to demonstrate compliance was October 5, 2004. We use the latter as the treatment
date in our analysis.30 Since many relevant aspects of the trading environment changed
around this time as well (e.g., the liquidity of the market increased, and many new sophis-
ticated investors entered the market) we need to compare cross-trades to a control group
of trades that are at least as likely as cross-trades to be a↵ected by increasing liquidity in
the markets post-2004 but are unlikely (or significantly less likely) to be directly a↵ected by
Rule 38a-1 and Rule 206(4)-7. Therefore, we compare the e↵ect of the introduction of new
regulation on the pricing of cross-trades (treatment group) with that on open market trades
(control group). Our analysis resembles a di↵erence-in-di↵erence in which only cross-trades
receive the treatment in October 2004. The e↵ect of internal governance on cross-trading
activity has so far never been explored and we believe it represents an interesting result in
itself.
B.2. The E↵ect of Increased Supervision on Cross-Trades
Figure 1 shows clearly that cross-trades and open market trades display a parallel trend before
the regulatory shock. However, the new regulation strongly a↵ects the execution shortfall
of cross-trades, while leaving the execution shortfall of open market trades unaltered. In
30Using as the treatment date February 5, 2004 provides us with quantitatively weaker but qualitativelyanalogous results.
21
particular, the execution shortfall of cross-trades is higher than open market trades before
the compliance date (see vertical line) and lower afterwards. Figure 2 shows the fraction of
cross-trades out of all institutional trades. In particular, the percentage of cross-trades starts
to decrease at the onset of the late trading scandal and drops permanently after the funds
had to comply with the new regulation. Overall, cross-trading activity went from peaks of 6%
of the dollar volume traded to less than 1% on average after the new rules were introduced.
Table III shows the e↵ect of tighter regulation on Execution Short f all in a multivariate
framework. Our specification includes CT Dummy; Post Regulation, i.e., a dummy variable
capturing the e↵ect of general changes in trading conditions after 2004; and the interaction
between Post Regulation and CT Dummy (Post Regulation is not included independently in
specifications (2) to (5) since it would be collinear with the time dummies). The control
group consists of open market trades that should be less (or not at all) a↵ected by the
change in regulation triggered by the late trading scandal. The coe�cient of the interacted
variables (CT Dummy x Post Regulation) captures the marginal e↵ect of the new regulation
on Execution Short f all for cross-trades (i.e., the e↵ect of the treatment). Post Regulation and
the time dummies capture the e↵ect of a general increase in market liquidity in the last part
of the sample.
Our results indicate that tighter regulation had a major e↵ect on the pricing of cross-
trades: Execution Short f all dropped by 59 bps almost immediately after the compliance
date, falling below that of open market trades (the result is significant at the 1% level). This
finding suggests that poor governance before the late trading scandal played a significant role
in determining a higher Execution Short f all for cross-trades. Overall, results in this section
are consistent with a causal relation between cross-trading and mispricing. Importantly,
while the execution shortfall of cross-trades does not exceed that of open market trades after
2004, the remaining deviation from benchmark prices may still be enough to arbitrarily shift
performance even if probably to a lesser extent. To find execution shortfalls systematically
higher than that of open market trades in the presence of tight supervision would be unlikely.
22
C. Stock Characteristics, Market Conditions, and Backdating
This section examines which cross-trades are more likely to present greater deviations from
benchmark prices. In our specification, we interact CT Dummy with a battery of stock-level
characteristics and market-level conditions. The stock-level variables are Illiquidity, 1/Price,
Bid-Ask Spread, Beta, and Volatility. Beta is the beta of the stock estimated using the Capital
Asset Pricing Model (CAPM), all other variables are described above (see Section III.A). The
market-level variables are the volatility index (V IX), the NBER recession indicator (NBER),
proxies of macroeconomic and financial uncertainty31 (Macro Uncert. and Fin. Uncert.), the
cross-sectional return dispersion in the day preceding the trade (CS Vol.), and the return
of the market in the previous month (Mkt Return). We control in our analysis for stock
characteristics non-interacted and time, stock, and family fixed e↵ects.
Results reported in Table IV indicate that the mispricing of cross-trades is more signif-
icant for highly illiquid32 and volatile stocks. This is not surprising since these stocks o↵er
more discretion on how to price transactions given that they incorporate higher asymmetric
information and have lower trading volume. One could argue that, by construction, highly
illiquid and volatile securities exhibit higher deviation from benchmark prices. Yet, includ-
ing time-varying stock characteristics non-interacted in all our specifications accounts for
this possibility (non-interacted stock and trade level variables are always included in Table
IV even though the coe�cients are not explicitly reported to save space). Goncalves-Pinto
and Sotes-Paladino (2015) posit that the main reason why funds cross-trade is the reduced
transaction cost when trading illiquid securities, we however show that institutions pay a sig-
nificantly higher cost when they cross-trade illiquid securities compared to when they trade
the same illiquid security in the open market.
In columns (4) and (5), we exclude time fixed e↵ects from our model in order to test
31We use proxies of macroeconomic and financial uncertainty from Jurado, Ludvigson, and Ng (2015).The authors derive model-free measures of uncertainty aggregating the h-step-ahead forecast error of severalfinancial and economic series.
321 over Price presents the “wrong” sign due to the high correlation with the other proxies of illiquidity.
23
whether market-wide conditions a↵ect how cross-trades are priced.33 We find that cross-
trades are more mispriced in times of uncertainty. Interestingly, most of the mispricing of
cross-trades appears to be unrelated to time-series volatility and positively related to measures
of asymmetric information in the markets (i.e., Jurado, Ludvigson, and Ng (2015)’s proxy
of financial uncertainty, measures of cross-sectional return dispersion, and illiquidity). This
is overall consistent with the hypothesis that institutions protect their top funds in periods
of high uncertainty o↵ering additional compensation to hold illiquid/di�cult to price assets.
Interestingly, the coe�cient of CT Dummy x Macro Uncertainty is negative, suggesting that,
in this case, cross-trades benefit the investors diminishing the cost of macroeconomic uncer-
tainty. In unreported results, we find that the coe�cient of CT Dummy x Macro Uncertainty
was positive before the 2004 regulation was introduced and turns negative afterwards. This
result supports the hypothesis that careful regulatory scrutiny may change dramatically how
cross-trades are used.
Additionally, we test whether cross-trades were backdated. When the main purpose of
cross-trades is to reallocate performance between counterparties, the best strategy from a
family perspective would be to arbitrarily set cross-trades ex post to the price of the day that
would have shifted the highest performance (i.e., the highest/lowest of the day). We test this
hypothesis estimating a logit model in which the dependent variable assumes a value of one
if the execution price is either the highest or the lowest of the day and we regress it on our
cross-trade dummy and controls. Whether the price was actually the highest or the lowest
is indi↵erent for our purpose since the party that is expected to gain from the transaction
can benefit in both cases (selling at the highest or buying at the lowest). Importantly, we
cannot include stock, family, and time fixed e↵ects contemporaneously in this specification
because the estimation of a non-linear model becomes infeasible with a very large number
of dummies. To limit our sample size we consider only families that cross-trade at least
once and to simplify the computation we include only family fixed e↵ects. In the Appendix,
33Also non-interacted market-wide variables and time-varying stock controls are included but the coe�-cients are not reported to save space.
24
we report analogous results estimating a linear probability model keeping in our sample all
observations and including time, family, and stock fixed e↵ects (see Table A.III).
Results reported in Table V suggest that some cross-trades may in fact have been back-
dated.34 Our estimated coe�cient indicates that cross-trades are 1.7% more likely to be
executed exactly at the highest/lowest price of the day (marginal probabilities are reported).
It is however certainly possible that traders choose to cross-trade when prices in the market
are extreme. Therefore, as in the previous section, we use the 2004 regulatory change as an
exogenous shock to improve our identification. We interact CT Dummy with Post Regulation
to assess whether cross-trades became less likely to be executed at the highest/lowest price af-
ter the new regulation was passed. Our findings are consistent with a causal interpretation of
our results: after 2004 cross-trades became 1.2% less likely to be executed at extreme prices.
The inclusion of open market trades rules out the possibility that market wide changes in
the trading environment after 2004 are driving our results (as they should also be a↵ected).
D. The Cross-Section of Cross-Trades
This section investigates how fund family characteristics a↵ected the pricing of cross-trades.
In general, family characteristics should not be correlated with the pricing of cross-trades.
However, if cross-trades were used to shift performance, we may find that proxies for weak
governance and high incentive to reallocate performance are correlated with the execution
shortfall. Importantly, to increase the power of our tests in this section only cross-trades are
kept. Since the number of cross-trades in the full ANcerno sample is relatively limited (we
have 738,476 cross-trades), we do not need to draw a random sample to conduct our analysis
but we can simply exclude all open-market trades from our sample. An alternative approach
would be to interact our CT Dummy with family characteristics and run our regressions on
a random extraction from ANcerno including both cross-trades and open market trades (as
34The issue of misreporting of execution times in ANcerno is arguably more relevant for this part of theanalysis since ANcerno arbitrarily sets missing time entries at the open/end of the trading day. To rule outthe possibility that this could influence our result, we replicate our analysis dropping all trades executedexactly at the opening or closing price of the day. Results are analogous.
25
we did in the previous sections). However, this would further limit the number of fund
families included in our analysis. Therefore, we decided to keep all cross-trades and to test
whether the cross-trades executed in some families are priced di↵erently from the cross-trades
executed in other families.
In our analysis, we focus on proxies of internal governance (Weak Governance)35, size of
internal markets (Siblings) and the incentive for performance redistribution (Fees Gap and
Expense Ratio Dispersion). The choice of these variables is motivated by the previous liter-
ature. In particular, Dimmock and Gerken (2012) argue that institutions that infringed the
law in the past are more likely to do it again and Massa (2003) suggests that fund prolif-
eration might be used opportunistically to attract flows. Additionally, theoretical models of
internal capital markets support the view that high heterogeneity in the importance of divi-
sions within multi-division companies leads to the reallocation of resources either to subsidize
weaker units or to support the stronger ones (Stein (1997), Stein and Scharfstein (2000)).
Similarly, Gaspar, Massa, and Matos (2006) and Chaudhuri, Ivkovich, and Trzcinka (2012)
argue that an asymmetry of “products” creates higher incentive to reallocate performance to
successful funds and powerful clients. On the contrary, in homogenous families in which all
funds have the same importance the incentive to shift performance is lower. In particular,
fund complexes have a strong incentive to move performance from the cheapest funds to the
funds charging the highest fees since outperformers attract disproportionate flows (Chevalier
and Ellison (1997), Sirri and Tufano (1998), Agarwal, Gay, and Ling (2014)).
We include in our regressions also the total asset size of the asset manager (Family Size)
to distinguish the e↵ect of large internal markets, in which many funds can cross-trade, from
the e↵ect of the size of the asset manager (as family size is a known predictor of fund returns,
see Chen, Hong, Huan, and Kubik (2004).) However, the high correlation of Family Size and
35A dummy equal to one for families investigated by the SEC for practices potentially harming investorsand zero otherwise, our approach follows Ferris and Yan (2009) and is motivated by Dimmock and Gerken(2012). In particular Dimmock and Gerken (2012) show that institutions that infringed the law in the pastwere more likely to do it again. We conjecture that having been engaged in suspicious practices, investigatedfamilies are on average more likely to lack the necessary control mechanisms to detect and avoid illegalcross-trading activity.
26
Siblings potentially creates problems due to multicollinearity concerns when we include both
in our specification at the same time. Therefore, we also present results obtained including
only one variable at the time. We also include cross-sectional return dispersion in the previous
month to exclude that our result just captures heterogeneity in fund performance unrelated
to cross-trading activity.
Our results reported in Table VI indicate that Weak Governance families display a 27
basis points higher Execution Short f all for cross-trades. Additionally, a standard deviation
increase in the number of siblings increases the Execution Short f all by 15 basis points, while
an increase of one standard deviation in Fees Gap boosts Execution Short f all by 10 basis
points. Results for the standard deviation of the expense ratio are similar (we do not include
both variables at the same time as they are highly correlated). We include stock and time
fixed e↵ects but not family fixed e↵ects since we are interested in estimating the e↵ect of
family-level variant and invariant characteristics. Overall, our results indicate than cross-
trades from weak governance institutions, with large internal markets, and high dispersion
in fees among siblings present higher Execution Short f all than the average cross-trade in our
sample. In particular, the e↵ect of fees dispersion on cross-trades pricing strongly points in the
direction of performance reallocation toward star funds as the incentive for “winner-picking”
strategies is stronger when some funds are significantly more valuable than others (Nanda,
Wang, and Zheng (2004)), while the incentive for subsidizing underperformers is probably
stronger when all funds have similar importance. Additional support for the“winner-picking”
hypothesis is provided in the next section.
Running our regressions only on the post regulation part of the sample (Table VI, Column
(8)), we find that asset managers with weak governance, numerous a�liated funds and large
heterogeneity in fees still price cross-trades at a higher deviation from benchmark prices after
2004 (the negative coe�cient of Family Size is just due to the high correlation with Siblings).
Our results therefore suggest that opportunistic pricing might still occur (even though the ex-
ecution shortall of cross-trades is significantly lower). Interestingly, in unreported regressions
27
we find that the e↵ect of Fees Gap on Execution Short f all is driven by the second part of the
sample. We conjecture that the dramatic growth of cheap passive funds and index trackers
in the last 10 years increased the average dispersion and gap in fees, while o↵ering a large
supply of liquidity providers to star funds. Some evidence in this direction is provided in
the following section. A systematic investigation on the e↵ect of opportunistic performance
reallocation on the performance of passive funds and index trackers is however left for future
research.
IV. Star Funds, Cross-trading, and Performance
Shifting
We believe that the evidence from trades provided in the previous section constitutes the
most important and novel contribution of our paper. However, exploiting our identification of
cross-trades we are able to shed some additional light on the ongoing debate on the incentives
at the fund family level. A necessary caveat is in order, the structure of our data allows us to
identify with high certainty the fraction of cross-trading activity at the asset manager level
but not the exact identity of the trading counterparties. While this was not an issue in the
previous section (as our analysis was conducted at the trade level), when we explore the e↵ect
on family-level cross-trading on fund performance we are going to inevitably add significant
noise to our analysis.
A. Methodology
In this section, we investigate whether fund families used cross-trades to boost the perfor-
mance of star funds (see, e.g., Gaspar, Massa, and Matos (2006)) or subsidize the junk funds
(see, e.g., Schmidt and Goncalves-Pinto (2013)). Our hypotheses derive from the literature
that explores the incentives of multi-division companies to allocate scarce resources to the
most successful units (picking winners) versus the least successful ones (in a framework that
28
subsidizes the worst performers).
These two alternative hypotheses have opposite empirical predictions. According to the
winner-picking hypothesis, cross-trading should increase the gap in performance between star
and junk funds. Conversely, the subsidization hypothesis predicts that cross-trading reduces
the spread in their performance. Importantly, non-opportunistic cross-trading could decrease
trading costs and, hence, improve funds’ performance even in the case of non-opportunistic
cross-trading. However, it should not be systematically correlated with the di↵erence in
performance between star and junk funds.
Since we show that cross-trades are on average mispriced, performance must be trans-
ferred between trading counterparties and this should be reflected into fund returns (unless
the party who benefits from the cross-trade is random and deviations from benchmark prices
average each other out). Our empirical strategy therefore consists, first, in defining groups
of funds inside a family that we hypothesize are likely to benefit or su↵er from cross-trading
and, second, in testing whether the di↵erence in returns correlates with cross-trading activity
within the family.
Our approach relies on the methodology introduced in Gaspar, Massa, and Matos (2006).
Specifically, in our main tests we rank funds according to their monthly36 flows (see, e.g.,
Bhattacharya, Lee, and Pool (2013)). The reason for ranking funds according to their flows
is intuitive.37 Funds with outflows are liquidity demanders and funds with inflows are the
natural liquidity suppliers. On the one hand, under a subsidization strategy star funds can
buy securities at inflated prices from the liquidity-demanding funds thereby increasing the
performance of the junk funds at their expenses. On the other hand, under a winner-picking
strategy, star funds can buy securities at deflated prices from the liquidity-demanding funds
(that are likely to be shut down anyway), increasing their own performance.38
36We focus on monthly observations because we cannot compute f lows at the daily level.37Ranking funds on net fees gives however qualitatively similar results.38An alternative approach would be to sort funds on gross fees (i.e., asset under management x percentage
fees), since the remuneration of mutual funds almost entirely consists of fees on asset under management(Haslem (2010)). However, it is very di�cult to subsidize large funds using cross-trades since the amount ofperformance transferred would need to be large. Since we find most of the mispricing to be in illiquid stocks,
29
Having ranked the funds, we then sort them into terciles for each family.39 Funds in
the intermediate tercile are discarded. From the two extreme terciles we construct pairwise
combinations matching funds from the top tercile with funds in the bottom tercile, and
we compute the spread in their style-adjusted performance (four-factor alpha). In order to
control for style e↵ects we impose as an additional restriction that the two funds operate in
the same investment style.
For instance, consider a family having six funds with the same investment style and
assume that in month t the funds all have di↵erent flows. This implies a ranking from 1 to 6
and two funds in each tercile. For our analysis we discard the funds ranked third and fourth
and we build the return spread from the remaining funds. Specifically, the observations in
our final sample would be the di↵erence of performance between fund 5 and fund 1, fund 5
and fund 2, fund 6 and fund 1, fund 6 and fund 2.
To understand how cross-trading shifts performance across siblings, we regress the spread
in performance between funds in the top tercile and bottom tercile on the percentage of cross-
trading activity, controlling for the invariant di↵erence in performance between the two funds,
family characteristics and observable di↵erences between the two funds. Formally:
rStari,t � rJunk
j,t = b(CT % f ,t)+G0Xi, j,t + gt + gi, j + ei, j,t , (4)
where rStari,t is the raw performance (or four-factor alpha) of star fund i in month t and
rJunkj,t is the raw performance (or four-factor alpha) of junk fund j in month t, provided that
both funds belong to the same fund family f and have the same investment style. CT % f ,t
is the percentage of cross-trading activity in family f where (i, j) 2 f . Xi, j,t is a vector of
fund/family-level controls accounting for observable di↵erences among the two funds (e.g.,
we recognize that it would be highly unlikely to boost the performance of large funds using such transactions.Hence, we focus on the “hot” funds, i.e., funds that attract the most new money within the family. To favorsuch funds makes economically sense as flows respond disproportionally to positive performance and havepositive spillovers to the rest of the family (Sirri and Tufano (1998), Nanda, Wang, and Zheng (2004), Basak,Pavlova, and Shapiro (2007)).
39Using quintiles yields similar results.
30
the di↵erence in funds’ size), gt and gi, j are time and fund pair fixed e↵ects. Fund pair fixed
e↵ects capture the invariant di↵erences among the funds, e.g., if the star fund manager is
on average more skilled than the junk fund manager.40 As an alternative specification we
control for family instead of fund pair fixed e↵ects. We do not include both because family
dummies are collinear to fund pair dummies.
The average spread = rStari,t � rJunk
j,t in our sample is positive (0.84% monthly risk-adjusted
return) since on average funds with higher flows (fees) outperform funds with lower flows
(fees). However, under the null hypothesis of no strategic interaction, we should expect to
find a correlation non-statistically di↵erent from zero between the spread in performance and
CT % f ,t , i.e., H0 : b = 0. Under the winner-picking hypothesis we should expect a positive
correlation between the spread in performance and CT % f ,t (i.e., cross-trading increases the
performance of star funds at the expense of the junk siblings), that is, H1 : b > 0. Under
the cross-subsidization hypothesis, we should expect a negative coe�cient (i.e., families shift
performance from star to junk funds, shrinking their performance gap), H2 : b < 0.
B. Winner-picking versus Subsidization of Junk Funds
In Table VII we investigate the e↵ect of cross-trading activity on the performance spread
between star and junk funds. We report results for the spread in style-adjusted returns41
(Columns 1-4) and for the spread in four-factor alphas (Columns 5-8). All of our regressions
include time fixed e↵ects and either fund pair or family fixed e↵ects. Errors are clustered
at the time level.42 It is important to stress that our proxy of cross-trading activity, CT %,
is at the family level, therefore our measure is likely to contain significant noise. The sign
and the coe�cient of b should however provide information on the direction of performance
40Skill might however be time-varying (see, e.g., Kacperczyk, Nieuwerburgh, and Veldkamp (2014)). Thisis not a concern under the assumption that cross-trading activity is unrelated to skill.
41Subtracting the return of a fund to the return of another fund having the same investment style we“clean” our measure of performance from the e↵ect of style.
42The di↵erence in performance between funds should be uncorrelated over time because there is no evidenceof persistence in performance (see, e.g., Carhart (1997), Frazzini and Lamont (2008), and Lou (2012).)Con-sistently, we find that clustering errors at the fund pair level does not change our results.
31
reallocation.
We find that the relation between CT % and the spread in returns is positive and strongly
significant (see Table VII). This result suggests that cross-trading activity widens the gap in
performance between star and junk funds.43 Overall, this empirical finding is consistent with
the winner-picking hypothesis and inconsistent with the cross-subsidization hypothesis.44 We
cannot however exclude that in some cases cross-subsidization of funds hit by redemptions
occurs. We would actually expect this to happen in a few cases, especially when flagship
funds are under significant pressure because of redemptions. However, our results indicate
that this does not occur on average.
In Columns 3, 4, 7, and 8 we also include a number of fund-level and family controls.
Specifically, to ensure that our results are not driven by di↵erences in the characteristics
between the two funds, we include their size di↵erence (DSize), their previous month return
di↵erence (DPastReturns), their previous month flow di↵erence (DPastFlow), and the di↵er-
ence in contemporaneous flows (DFlow).45 We also include Family Size to account for the
positive correlation between cross-trading activity and the size of the mutual fund complex,
and Return Dispersion to make sure that our results are not driven by ex ante heterogeneity
in fund returns at the family level.
Our estimates suggest that one standard deviation increase in monthly cross-trading ac-
tivity increased by about 24 basis points the risk-adjusted performance gap between junk
and star funds (see Table VII, Column 8). Considering families in which there is no cross-
trading activity as the control group, our back-of-the-envelope calculations suggest that star
43To make sure that our result is not driven by unobservable di↵erences between families that do and do notcross-trade, we replicate the same analysis dropping all observations where CT % = 0. Results are unchanged(see Appendix).
44A natural question that arises is why the manager that get penalized from the cross-trade should engagein it. We can conjecture three explanations that we cannot however test with our data. First, the manager ofthe two funds that are cross-trading may actually be the same, hence she would simply boost the performanceof her top fund. Second, it is possible that a fund that is about to get closed is penalized to the benefit of itssiblings. Third, the most heavily penalized funds might be passive funds and index trackers.
45We control for contemporaneous flows because when we sort funds on flows we mechanically generate aspread in performance. This should not be a problem as long as cross-trading is not a↵ected by flows. Tomitigate any potential endogeneity concern we therefore control for contemporaneous flow and we exploit thechange in the regulatory environment described in Section IV.
32
funds boosted their risk-adjusted performance by 1.7% annually at the expense of junk funds,
assuming that cross-trading funds have equal size and performance is shared equally.46 Ad-
ditionally, Table VIII shows that the inflow funds that benefited from cross-trading activity
were only those that charged higher than median fees - the coe�cient of CT % x High Fees
is positive and significant, while that of CT % becomes statistically non-di↵erent from zero.
Overall, it appears that the subset of funds that benefited from cross-trading includes only
those that were most valuable from a family perspective.
Our results so far suggest that fund families used cross-trading to shift performance from
junk funds to star siblings. However, reverse causality and omitted variable bias are a concern
also in this setting. For instance, fund families with a higher spread in performance may cross-
trade more or omitted factors may drive both cross-trading and performance. Again, we use
the regulatory change that followed the late trading scandal to establish causality. In Table
IX we add to our main specification an interaction variable between CT % and Post, i.e., a
dummy variable taking a value of one after the compliance date was reached and taking a
value of zero otherwise. Post captures the e↵ect of changes in the trading environment in
the post-regulation sample that are unrelated to cross-trading activity. We do not include
the dummy Post non-interacted with CT % in specifications 2-4 and 6-8 because the variable
would be completely spanned by the time fixed e↵ects. Contrary to our previous specification,
we should expect to find the coe�cient of the interaction between CT % and Post (b) to be
negative in the case of winner-picking behavior (b < 0) as the new regulation should reduce
the gap in the performance between star and junk funds that was due to cross-trading.
Conversely, we should find b > 0 if funds used cross transactions to support junk funds, i.e.,
the performance gap should have been artificially low before 2004 and should now increase.
b = 0 should be expected in case the new regulation did not have any impact on the e↵ect
of cross-trading on performance. The inclusion of time and family dummies rules out the
possibility that the e↵ect is driven by changes in the market environment or by unaccounted
46The marginal e↵ect of a cross-trading dummy is 0.28% (see Table A.IV in the Appendix). If we assumethat performance is shared equally each counterparty gains or loses 0.14% per month, i.e., 1.7% annually.
33
family characteristics.
Our results indicate that the new regulation was on average e↵ective in eliminating the
impact of cross-trading activity on the spread in performance between star and junk funds.
The marginal e↵ect of cross-trading on performance is almost completely balanced out by the
negative e↵ect of the new regulation. Overall, measuring cross-trading activity using actual
cross-trades instead of opposite side transactions, we provide support for the winner picking
hypothesis thereby confirming the evidence from opposite trades provided in Gaspar, Massa,
and Matos (2006). Conversely, we rule out the hypothesis of systematic cross-subsidization
of distressed funds (see, e.g., Schmidt and Goncalves-Pinto (2013)). As our results contain
significant noise, given that cross-trades are computed at a family level, we are unable in
this section to assess whether cross-trades still shift performance to some funds after the new
regulation was introduced.
V. Further Results and Robustness
This section provides additional results and robustness checks.
A. Alternative Benchmark Prices
Most of the results provided in the paper use the price of the stock in the market at the
moment of the execution as the main benchmark as this seems to be the closest to what Rule
17a-7 of the U.S. Investment Company Act requires. However, in this section we show that
our results are analogous choosing di↵erent benchmarks. As a first alternative benchmark, we
replicate our trade-level analysis using the volume-weighted average price of the day instead
of the price at the moment of the execution. Formally:
Execution Short f all j,i,t =|Pj,i,t �VWAPi,d|
VWAPi,d, (5)
where Pj,i,t is the execution price of trade j, in stock i, at execution time t of day d; while
34
VWAPi,d is the volume-weighted average price for stock i in day d when trade j is executed.
Results for the regression of this alternative measure of Execution Shortfall on CT Dummy
are reported in Table X, Panel A. The results are qualitatively similar to those reported in
Table II (i.e., using the market price at the moment of the execution as the benchmark price).
Results obtained replicating the other tests in the paper using VWAPi,d as main benchmark
are also qualitatively similar and are therefore unreported. We have chosen not to present the
results obtained using volume-weighted average price benchmark as main results in the paper
as the use of VWAP has potentially a few shortcomings (see Hasbrouck (2007), p. 148). For
example, if a trade accounts for a large proportion of the daily volume, the weighted average
execution price of the trade is likely to coincide with the VWAP.
As a second benchmark, we replicate our analysis using the opening price of the day. To
make sure that our results are not driven by misreporting (some trades from ANcerno are
arbitrarily set at the open price of the day), we exclude the trades executed exactly at the
opening price. Therefore, we compute the execution shortfall as follows:
Execution Short f all j,i,t =|Pj,i,t �Openi,d|
Openi,d, (6)
where Openi,d is the opening price for stock i in day d. Results are reported in Table X,
Panel B and are unchanged.
B. Cross-trades and Commissions
Our previous sections show that cross-trades were significantly mispriced (we estimate a
marginal e↵ect of cross-trades on Execution Short f all of 0.18%) and likely to reallocate per-
formance from trading counterparties. Yet we also show that commissions paid on each dollar
worth of cross-trading are significantly lower (around 10 basis points less than open market
trades, see Table I). Is the di↵erence in execution shortfall negligible after taking commis-
sions into account? We replicate our analysis adding percentage commissions to the execution
shortfall. Results reported in Table XI show that cross-trades exhibit a 0.12% higher execu-
35
tion shortfall than open market trades a f ter commissions are taken into account. Overall, our
results indicate that the e↵ect of cross-trades on performance was economically significant.
VI. Conclusion
In this paper, we exploit institutional trade-level data provided by ANcerno to shed light on
cross-trading practices. In general trading in opaque and lower regulated venues has increased
in recent years. To measure cross-trades, we look for pair of trades coming from funds belong-
ing to the same fund family, in the same stock, involving the exact same quantity of shares
traded, and sharing the same execution day, time, and price. Previous literature focuses on
measures of cross-trading inferred by opposite side trades (often of di↵erent volumes) with
the only requirement of occurring in the same quarter, thereby significantly misrepresenting
real cross-trading activity.
Using our precise measure, we show that cross-trades in a sample of trades that cover
the 1999-2010 period exhibit an execution shortfall that is 0.18% higher than open market
trades, 4.5 times the average percentage bid-ask spread in our sample. Additionally, we show
that the execution price of the cross-trades appears to be sometimes set ex post to the highest
or lowest price of the day. Finally, we show that the execution shortfall of cross-trade was
substantial in illiquid and highly volatile stocks, in uncertain times, and in the presence of
weak supervision or governance.
We exploit an exogenous shock to industry regulation to rule out alternative explanations
based on reverse causality, illiquidity, or changing trading conditions. We find that both the
incentive to cross-trade and the severity of the mispricing diminished drastically when regula-
tory scrutiny increased. Overall, we o↵er support to the hypothesis that star funds benefited
from cross-trading at the expense of junk funds. This strategy had relevant implications for
fund ranking, fund selection, and fund manager evaluation. Our results suggest that fund
alphas potentially misrepresent the real ability of fund managers to create value for their
36
investors.
37
References
Agarwal, Vikas, Gerald D Gay, and Leng Ling, 2014, Window dressing in mutual funds,
forthcoming, Review of Financial Studies.
Anand, Amber, Paul Irvine, Andy Puckett, and Kumar Venkataraman, 2012, Performance of
institutional trading desks: An analysis of persistence in trading costs, Review of Financial
Studies 25, 557–598.
Anton, Miguel, and Christopher Polk, 2014, Connected stocks, The Journal of Finance 69,
1099–1127.
Basak, Suleyman, Anna Pavlova, and Alexander Shapiro, 2007, Optimal asset allocation and
risk shifting in money management, Review of Financial Studies 20, 1583–1621.
Ben-David, Itzhak, and David Hirshleifer, 2012, Are investors really reluctant to realize their
losses? trading responses to past returns and the disposition e↵ect, Review of Financial
Studies 25, 2485–2532.
Berkowitz, Stephen A, Dennis E Logue, and Eugene A Noser, 1988, The total cost of trans-
actions on the NYSE, The Journal of Finance 43, 97–112.
Bhattacharya, Utpal, Jung H Lee, and Veronika K Pool, 2013, Conflicting family values in
mutual fund families, The Journal of Finance 68, 173–200.
Brown, David P, and Youchang Wu, 2015, Mutual fund flows and cross-fund learning within
families, The Journal of Finance.
Brown, Keith C, W Van Harlow, and Laura T Starks, 1996, Of tournaments and temptations:
An analysis of managerial incentives in the mutual fund industry, The Journal of Finance
51, 85–110.
Busse, Je↵rey A, Tarun Chordia, Lei Jiang, and Yuehua Tang, 2015, Mutual fund trading
costs and diseconomies of scale, Working paper.
38
Carhart, Mark, 1997, Persistence in mutual fund performance, The Journal of Finance 57,
28–57.
Casavecchia, Lorenzo, and Ashish Tiwari, 2015, Cross trading by investment advisers: Im-
plications for mutual fund performance, Journal of Financial Intermediation.
Chaudhuri, Ranadeb, Zoran Ivkovich, and Charles Trzcinka, 2012, Strategic performance
allocation in institutional asset management firms: Behold the power of stars and dominant
clients, Working paper.
Chemmanur, Thomas J, Shan He, and Gang Hu, 2009, The role of institutional investors in
seasoned equity o↵erings, Journal of Financial Economics 94, 384–411.
Chen, Joseph, Harrison Hong, Ming Huan, and Je↵rey D. Kubik, 2004, Does fund size erode
mutual fund performance? the role of liquidity and organization, American Economic
Review 94, 1276–1302.
Chen, Joseph, Harrison Hong, Wenxi Jiang, and Je↵rey D Kubik, 2013, Outsourcing mutual
fund management: firm boundaries, incentives, and performance, The Journal of Finance
68, 523–558.
Chevalier, Judith, and Glenn Ellison, 1997, Risk taking by mutual funds as a response to
incentives, Journal of Political Economy 105, 1167–1200.
Chuprinin, Oleg, Massimo Massa, and David Schumacher, 2015, Outsourcing in the interna-
tional mutual fund industry: An equilibrium view, forthcoming, The Journal of Finance.
Comerton-Forde, Carole, and Talis J Putnins, 2015, Dark trading and price discovery, Journal
of Financial Economics 118, 70–92.
Coval, Joshua, and Erik Sta↵ord, 2007, Asset fire sales (and purchases) in equity markets,
Journal of Financial Economics 86, 479–512.
39
Dimmock, Stephen G, and William C Gerken, 2012, Predicting fraud by investment man-
agers, Journal of Financial Economics 105, 153–173.
Elton, Edwin J, Martin J Gruber, and T Clifton Green, 2007, The impact of mutual fund
family membership on investor risk, Journal of Financial and Quantitative Analysis 42,
257.
Evans, Richard B, 2010, Mutual fund incubation, The Journal of Finance 65, 1581–1611.
Ferreira, Miguel A, Aneel Keswani, Antonio F Miguel, and Sofia B Ramos, 2012, The deter-
minants of mutual fund performance: A cross-country study, Review of Finance p. rfs013.
Ferris, Stephen P, and Xuemin Sterling Yan, 2009, Agency costs, governance, and organiza-
tional forms: Evidence from the mutual fund industry, Journal of Banking & Finance 33,
619–626.
Frazzini, Andrea, and Owen A Lamont, 2008, Dumb money: Mutual fund flows and the
cross-section of stock returns, Journal of Financial Economics 88, 299–322.
Gaspar, Jose-Miguel, Massimo Massa, and Pedro Matos, 2006, Favoritism in mutual fund
families? evidence on strategic cross-fund subsidization, The Journal of Finance 61, 73–
104.
Goncalves-Pinto, Luis, and Juan Sotes-Paladino, 2015, The invisible hand of internal markets
in mutual fund families, Working Paper, National University of Singapore.
Guedj, Ilan, and Jannette Papastaikoudi, 2008, Can mutual fund families a↵ect the perfor-
mance of their funds, Working paper, University of Texas at Austin.
Hasbrouck, Joel, 2007, Empirical market microstructure: The institutions, economics, and
econo- metrics of securities trading (Oxford University Press).
Haslem, John A, 2010, Mutual Funds: Portfolio structures, analysis, management, and stew-
ardship (Wiley).
40
Hu, Gang, 2009, Measures of implicit trading costs and buy–sell asymmetry, Journal of
Financial Markets 12, 418–437.
Huang, Jennifer, Clemens Sialm, and Hanjiang Zhang, 2011, Risk shifting and mutual fund
performance, Review of Financial Studies.
Jurado, K, S Ludvigson, and S Ng, 2015, Forthcoming. measuring uncertainty, American
Economic Review.
Kacperczyk, Marcin, Stijn Van Nieuwerburgh, and Laura Veldkamp, 2014, Time-varying fund
manager skill, The Journal of Finance 69, 1455–1484.
Kacperczyk, Marcin, Clemens Sialm, and Lu Zheng, 2005, On the industry concentration of
actively managed equity mutual funds, The Journal of Finance 60, 1983–2011.
, 2008, Unobserved actions of mutual funds, Review of Financial Studies 21, 2379–
2416.
Kempf, Alexander, and Stefan Ruenzi, 2008, Tournaments in mutual-fund families, Review
of Financial Studies 21, 1013–1036.
, and Tanja Thiele, 2009, Employment risk, compensation incentives, and managerial
risk taking: Evidence from the mutual fund industry, Journal of Financial Economics 92,
92–108.
Lou, Dong, 2012, A flow-based explanation for return predictability, Review of Financial
Studies 25, 3457–3489.
Madhavan, Ananth N, 2002, Vwap strategies, Trading 2002, 32–39.
Massa, Massimo, 2003, How do family strategies a↵ect fund performance? when performance-
maximization is not the only game in town, Journal of Financial Economics 67, 249–304.
41
, and Rajdeep Patgiri, 2009, Incentives and mutual fund performance: higher perfor-
mance or just higher risk taking?, Review of Financial Studies 22, 1777–1815.
McCabe, Patrick E, 2009, The economics of the mutual fund trading scandal, Working paper,
Federal Reserve Board.
Nanda, Vikram, Z. Jay Wang, and Lu Zheng, 2004, Family values and the star phenomenon:
Strategies of mutual fund families, Review of Financial Studies 17, 668–698.
Pollet, J.M., and M. Wilson, 2008, How does size a↵ect mutual fund behavior?, The Journal
of Finance 63, 2941–2969.
Puckett, Andy, and Xuemin Sterling Yan, 2011, The interim trading skills of institutional
investors, The Journal of Finance 66, 601–633.
Schmidt, Breno, and Luis Goncalves-Pinto, 2013, Co-insurance in mutual fund families,
Working Paper, Emory University.
Sirri, Erik R., and Peter Tufano, 1998, Costly search and mutual fund flows, The Journal of
Finance 53, 1589–1622.
Stein, Jeremy, and David S Scharfstein, 2000, The dark side of internal capital markets:
divisional rent-seeking and ine�cient investment, Journal of Finance 55.
Stein, Jeremy C, 1997, Internal capital markets and the competition for corporate resources,
The Journal of Finance 52, 111–133.
Zhu, Haoxiang, 2014, Do dark pools harm price discovery?, Review of Financial Studies 27,
747–789.
Zitzewitz, Eric, 2006, How widespread was late trading in mutual funds?, American Economic
Review 96, 284–289.
42
0"
0.005"
0.01"
0.015"
0.02"
0.025"
1999m2"
1999m5"
1999m8"
1999m11"
2000m2"
2000m5"
2000m8"
2000m11"
2001m2"
2001m5"
2001m8"
2001m11"
2002m2"
2002m5"
2002m8"
2002m11"
2003m2"
2003m5"
2003m8"
2003m11"
2004m2"
2004m5"
2004m8"
2004m11"
2005m2"
2005m5"
2005m8"
2005m11"
2006m2"
2006m5"
2006m8"
2006m11"
2007m2"
2007m5"
2007m8"
2007m11"
2008m2"
2008m5"
2008m8"
2008m11"
2009m2"
2009m5"
2009m8"
2009m11"
2010m2"
2010m5"
2010m8"
2010m11"
Open"Mkt"Trades" Cross<Trades"
Figure 1: Execution Shortfall over time for cross-trades and open market trades. Executionshort f all is defined as follows: Execution Short f all j,i,t =
|Pj,i,t�Pi,t |Pi,t
, where Pj,i,t is the execution price of trade j, instock i, at execution time t; while Pi,t is the price of stock i in the market at time t. We present results obtainedcomputing three-month moving averages in order to smooth the series. Cross-trades are defined as indicatedin Section II.B. In connection with the investigation into illegal trading practices in the mutual fund industry,on September 3, 2003 New York Attorney General Eliot Spitzer announced the issuance of a complaint againstCanary Capital Partners LLC claiming that they had engaged in late trading. As a consequence rules 38a-1and 206(4)-7 and the amendments to rule 204-2 were introduced. Industry participants had to comply to thenew rules by October 5, 2004 (see vertical line).
43
0"
0.01"
0.02"
0.03"
0.04"
0.05"
0.06"
0.07"
1999m2"
1999m5"
1999m8"
1999m11"
2000m2"
2000m5"
2000m8"
2000m11"
2001m2"
2001m5"
2001m8"
2001m11"
2002m2"
2002m5"
2002m8"
2002m11"
2003m2"
2003m5"
2003m8"
2003m11"
2004m2"
2004m5"
2004m8"
2004m11"
2005m2"
2005m5"
2005m8"
2005m11"
2006m2"
2006m5"
2006m8"
2006m11"
2007m2"
2007m5"
2007m8"
2007m11"
2008m2"
2008m5"
2008m8"
2008m11"
2009m2"
2009m5"
2009m8"
2009m11"
2010m2"
2010m5"
2010m8"
2010m11"
Figure 2: Percentage of cross-trading activity over time. Cross-trading activity is computed as themonthly dollar amount of cross-trading over the monthly dollar amount of total trading. The three-monthmoving average is plotted in order to smooth the series. Cross-trades are defined as indicated in Section II.B.In connection with the investigation into illegal trading practices in the mutual fund industry, on September3, 2003 New York Attorney General Eliot Spitzer announced the issuance of a complaint against CanaryCapital Partners LLC claiming that they had engaged in late trading. As a consequence rules 38a-1 and206(4)-7 and the amendments to rule 204-2 were introduced. Industry participants had to comply to the newrules by October 5, 2004 (see vertical line).
44
Table I: Summary Statistics
This table provides summary statistics for our sample (Panel A) and correlations among the main variables(Panel B). The average values reported are obtained extracting a 1% random sample of trades withoutreplacement from ANcerno. Cross-trades are defined as trades that occur in the same stock, the samequantity, the same price, on the same day and time but display opposite side as at least one other tradereported by the same fund family. All other trades are defined as open market trades. Column (1) reportsthe number of observations available for each variable, Column (2) reports the average value of the variableirrespectively on whether a trade is crossed or not, Column (3) reports averages for open market trades only,Column (4) reports averages for cross-trades only, Column (5) reports the di↵erence between open markettrades and cross-trades (i.e., the di↵erence between Column (3) and Column (4)), Column (6) indicatest-statistics for a two-sided test on whether the di↵erence reported in Column (5) is statistically di↵erentfrom zero. Share Volume is the average share size of the trade; Dollar $Volume is the average size of the
trade in dollars; Execution Short f all is defined as follows: Execution Short f all j,i,t =|Pj,i,t�Pi,t |
Pi,t, where Pj,i,t is the
execution price of trade j, in stock i, at execution time t; while Pi,t is the price of stock i in the market attime t; Illiquidity is Amihud’s monthly illiquidity ratio computed from daily returns obtained from CRSP;Bid�Ask Spread is the di↵erence between the bid and the ask at the beginning of the month as reported fromCRSP; 1/Price is 1 over the opening price of the day; Market Equity Decile is the equity decile computed usingNYSE breakpoints; S&P500 Dummy equals one if a stock is included in the S&P500 index and zero otherwise;Stock Volatility is the within-month standard deviation of daily stock returns; Commissions($/share) is thedollar commission paid for a trade over share volume; Commissions($/$trade) is the dollar commission paidfor a trade over dollar trade volume.
Panel A: Sample StatisticsObservations Full Sample Open Trades Cross-Trades Di↵. t-stat.
(1) (2) (3) (4) (5) (6)
Share Volume 966,186 7,092 7,014 17,332 -10,318 -22.50Dollar Volume 966,186 21,5581 212,707 589,603 -376,896 -28.10Execution Price 966,186 42.58 42.57 44.17 -1.60 -0.23Execution Shortfall 965,711 0.0065 0.0064 0.0084 -0.0019 -16.29Illiquidity 966,186 0.0443 0.0445 0.0268 0.0177 0.090Bid-Ask Spread 966,186 0.0031 0.0031 0.0041 -0.0011 -14.42S&P500 Dummy 966,186 0.5153 0.5148 0.5817 -0.0668 -11.44Volatility 966,186 0.1133 0.1132 0.1270 -0.0139 -14.93Market Equity Decile 966,186 7.2198 7.2150 7.8377 -0.6226 -18.921/Price 966,186 0.0518 0.0518 0.0469 0.0049 2.88Commissions ($/share) 965,595 0.0243 0.0245 0.0016 0.0229 69.45Commissions ($/$trade) 965,595 0.0011 0.0011 0.0001 0.0010 8.040
Panel B: CorrelationsExecution S. S Volume B/M ME Bid-Ask 1/Price
Execution Shortfall 1.0000Share Volume 0.1194 1.0000B/M Dec 0.0097 0.0181 1.0000ME Dec -0.1208 -0.0135 -0.2669 1.0000Bid-Ask Spread 0.1815 0.1373 0.0729 -0.1476 1.00001/Price 0.1279 0.0743 0.1615 -0.2776 0.2822 1.0000
45
Table II: Do Cross-Trades Exhibit Higher Execution Shortfall?
This table reports OLS estimates obtained by regressing Execution Short f all on CT Dummy and controls.
Execution Short f all is defined as follows: Execution Short f all j,i,t =|Pj,i,t�Pi,t |
Pi,t, where Pj,i,t is the execution price
of trade j, in stock i, at execution time t; while Pi,t is the price of stock i in the market at time t. CT Dummyequals one if a trade is a cross-trade and equals zero when a trade is executed in the open market. Volume isthe share volume of the trade; Illiquidity is Amihud’s monthly illiquidity ratio computed from daily returnsobtained from CRSP; Bid �Ask Spread is the di↵erence between the bid and the ask at the beginning ofthe month as reported from CRSP; 1/Price is 1 over the opening price of the day; Market Equity Decile isthe equity decile computed using NYSE breakpoints; S&P500 Dummy equals one if a stock is included inthe S&P500 index and zero otherwise; Stock Volatility is the within-month standard deviation of daily stockreturns. Observations are at the trade level and are obtained by drawing a 1% random sample of trades fromANcerno without replacement. Stock, time, and family fixed e↵ects are included and errors are clustered atthe time level. The constant is included in all specifications but the coe�cient is not reported. ***, **, *indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Execution Shortfall
(1) (2) (3) (4) (5)
CT Dummy 0.0019*** 0.0018*** 0.0018*** 0.0019*** 0.0018***(5.44) (5.35) (5.35) (5.40) (5.37)
Volume 0.0002*** 0.0002*** 0.0002*** 0.0002***(12.00) (12.01) (11.57) (11.24)
Illiquidity 0.0402*** 0.0265*** 0.0287***(3.83) (4.63) (4.55)
Bid-Ask Spread -0.0057 -0.0041(-0.40) (-0.37)
1/Price 0.0037*** 0.0027***(4.04) (3.69)
Market Equity Decile -0.0001**(-2.00)
S&P 500 Dummy -0.0003***(-3.17)
Volatility 0.0195***(17.97)
Stock Fixed E↵ect Yes Yes Yes Yes YesFamily Fixed E↵ect Yes Yes Yes Yes YesTime Fixed E↵ect Yes Yes Yes Yes YesObservations 964,972 964,972 964,972 964,972 964,972R-squared 0.208 0.209 0.209 0.211 0.220
46
Table III: What was the Impact of Restrictive Regulation on the Pricing of the Cross-Trades?
This table reports OLS estimates obtained by regressing Execution Short f all on CT Dummy, Post Regulation,and controls. Execution Short f all is defined as follows: Execution Short f all j,i,t =
|Pj,i,t�Pi,t |Pi,t
, where Pj,i,t is the
execution price of trade j, in stock i, at execution time t; while Pi,t is the price of stock i in the market attime t. Post Regulation equals one for trades executed from October 2004 onwards and equals zero before ofthat; CT Dummy equals one if a trade is a cross-trade and equals zero when a trade is executed in the openmarket. Volume is the share volume of the trade; Illiquidity is Amihud’s monthly illiquidity ratio computedfrom daily returns obtained from CRSP; Bid �Ask Spread is the di↵erence between the bid and the ask atthe beginning of the month as reported from CRSP; 1/Price is 1 over the opening price of the day; MarketEquity Decile is the equity decile computed using NYSE breakpoints; S&P500 Dummy equals one if a stock isincluded in the S&P500 index and zero otherwise; Stock Volatility is the within-month standard deviation ofdaily stock returns. Observations are at the trade level and are obtained by drawing a 1% random sample oftrades from ANcerno without replacement. Stock, time, and family fixed e↵ects are included when specifiedand errors are clustered at the time level. The constant is included in all specifications but the coe�cient isnot reported. ***, **, * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Execution Shortfall
(1) (2) (3) (4) (5)
CT Dummy x Post Regulation -0.0066*** -0.0061*** -0.0059*** -0.0059*** -0.0059***(-16.30) (-15.74) (-15.65) (-15.79) (-15.67)
CT Dummy 0.0049*** 0.0048*** 0.0047*** 0.0047*** 0.0046***(15.41) (15.82) (15.47) (15.63) (15.52)
Post Regulation -0.0030***(-7.16)
Volume 0.0002*** 0.0002*** 0.0002***(11.89) (11.44) (11.11)
Illiquidity 0.0403*** 0.0265*** 0.0287***(3.83) (4.62) (4.54)
Bid-Ask Spread -0.0051 -0.0036(-0.36) (-0.33)
1/Price 0.0037*** 0.0027***(4.03) (3.68)
Market Equity Decile -0.0001**(-2.03)
S&P500 Dummy -0.0003***(-3.12)
Volatility 0.0195***(17.96)
Stock Fixed E↵ect Yes Yes Yes Yes YesFamily Fixed E↵ect Yes Yes Yes Yes YesTime Fixed E↵ect No Yes Yes Yes YesObservations 964,972 964,972 964,972 964,972 964,972R-squared 0.161 0.209 0.210 0.211 0.220
47
Table IV: Which/When Cross-Trades are Mispriced?
This table reports OLS estimates obtained by regressing Execution Short f all on CT Dummy, interactions ofCT Dummy and stock and markets characteristics, and controls. Execution Short f all is defined as follows:
Execution Short f all j,i,t =|Pj,i,t�Pi,t |
Pi,t, where Pj,i,t is the execution price of trade j, in stock i, at execution time t;
while Pi,t is the price of stock i in the market at time t. CT Dummy equals one if a trade is a cross-trade andequals zero when a trade is executed in the open market. Volume is the share volume of the trade; Illiquidity isAmihud’s monthly illiquidity ratio computed from daily returns obtained from CRSP; Bid�Ask Spread is thedi↵erence between the bid and the ask at the beginning of the month as reported from CRSP; 1/Price is 1 overthe opening price of the day; Market Equity Decile is the equity decile computed using NYSE breakpoints;S&P500 Dummy equals one if a stock is included in the S&P500 index and zero otherwise; Stock Volatility isthe within-month standard deviation of daily stock returns. Beta is the stock market beta estimated assumingthe CAPM. V IX is the Volatility Index, NBER is a dummy variable that takes value one during crises andequals zero otherwise. Macro and Financial Uncertainty are from Jurado, Ludvigson, and Ng (2015); CS Volis the cross-sectional standard deviation of daily returns in the previous day, Mkt Return is cumulative stockmarket return in the previous month. All non-interacted variables are included but coe�cients arenot reported. Observations are at the trade level and are obtained by drawing a 1% random sample oftrades from ANcerno without replacement. Stock, time, and family fixed e↵ects are included and errors areclustered at the time level. The constant is included in all specifications but the coe�cient is not reported.***, **, * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Execution Shortfall
(1) (2) (3) (4) (5)
Stock Characteristics
CT Dummy x Illiquidity 0.2248*** 0.1633** 0.1416** 0.1813** 0.1930**(3.49) (2.38) (2.16) (2.19) (2.55)
CT Dummy x Bid-Ask Spread 0.1103*** 0.1122*** 0.0981*** -0.0282 -0.0414(4.60) (4.73) (3.53) (-0.83) (-1.05)
CT Dummy x 1/Price -0.0045 -0.0061** -0.0110*** -0.0093*** -0.0059**(-1.53) (-2.07) (-3.53) (-3.03) (-1.99)
CT Dummy x Beta 0.0003 0.0002 0.0001(1.14) (0.73) (0.20)
CT Dummy x Volatility 0.0145*** 0.0135*** 0.0118***(3.43) (3.30) (2.80)
Market Conditions
CT Dummy x VIX 0.0000 -0.0000(0.07) (-0.83)
CT Dummy x NBER 0.0010 0.0004(0.93) (0.35)
CT Dummy x Macro Uncert. -0.0351*** -0.0318***(-9.21) (-8.31)
CT Dummy x Fin. Uncert. 0.0095*** 0.0071***(4.50) (3.08)
CT Dummy x CS Vol. 0.0668***(2.83)
CT Dummy x Mkt Return 0.0003(0.04)
CT Dummy 0.0016*** 0.0017*** -0.0003 0.0157*** 0.0141***(4.61) (4.81) (-0.88) (7.08) (5.84)
Time-Varying Controls No Yes Yes Yes YesStock Fixed E↵ect Yes Yes Yes Yes YesFamily Fixed E↵ect Yes Yes Yes Yes YesTime Fixed E↵ect Yes Yes Yes No NoObservations 964,972 964,972 951,993 951,993 951,993R-squared 0.211 0.220 0.219 0.194 0.204
48
Table V: Are Cross-Trades Backdated?
This table reports logit estimates of the probability of a trade to be executed either at exactly the highest orat exactly the lowest price of the day(marginal probabilities are reported). CT Dummy equals one if a tradeis a cross-trade and equals zero when the trade is executed in the open market. Post Regulation equals onefor trades executed from October 2004 onwards and equals zero before of that; Volume is the share volume ofthe trade; Illiquidity is Amihud’s monthly illiquidity ratio computed from daily returns obtained from CRSP;Bid�Ask Spread is the di↵erence between the bid and the ask at the beginning of the month as reported fromCRSP; 1/Price is 1 over the opening price of the day; Market Equity Decile is the equity decile computed usingNYSE breakpoints; S&P500 Dummy equals one if a stock is included in the S&P500 index and zero otherwise;Stock Volatility is the within-month standard deviation of daily stock returns. Observations are at the tradelevel and are obtained by drawing a 1% random sample of trades from ANcerno without replacement. Onlyobservations from families that cross-trade at least once are included. Family fixed e↵ects are included anderrors are clustered at the time level. Results using the full specification model are presented in Table A.III.The constant is included in all specifications but the coe�cient is not reported. ***, **, * indicate statisticalsignificance at the 1%, 5%, and 10% level, respectively.
Highest/Lowest Price of the Day
(1)
CT Dummy x Post Regulation -0.0121***(-5.09)
CT Dummy 0.0172***(9.76)
Post Regulation -0.0071***(-8.29)
Volume -0.0023***(-18.97)
Illiquidity -0.0352(-1.56)
Bid-Ask Spread 0.4806***(12.96)
1/Price 0.0014***(2.13)
Market Equity Decile -0.0029***(-22.24)
S&P500 Dummy 0.0036***(5.57)
Volatility -0.0323***(-7.26)
Family Fixed E↵ect YesObservations 816,721Pseudo R2 0.12
49
Table
VI:
DoDi↵eren
tFund
FamiliesPrice
Cro
ss-T
rades
Di↵er
ently?
This
table
reports
OLSestimates
obtained
byregressing
Exe
cutio
nSh
ortf
allon
mutual
fundfamilylevelvariab
lesan
dcontrols.Im
portantly,only
cross-tra
des
are
included
.E
xecu
tion
Shor
tfal
lis
defined
asfollow
s:E
xecu
tion
Shor
tfal
l j,i,t=
|Pj,i,t�
P i,t|
P i,t
,where
P j,i,
tis
theexecution
price
oftrad
e
j,in
stock
i,at
execution
time
t;while
P i,tis
theprice
ofstock
iinthemarketat
time
t;W
eak
Gov
erna
nceequalson
eforfamiliesinvestigated
bythe
SEC
forpractices
potentially
harminginvestorsan
dequalszero
otherwise;
Sibl
ings
isthenaturallogof
thenu
mber
ofequityfundsin
thefundfamily;
Retu
rnD
ispe
rsio
nis
thelagged
mon
thly
cross-sectional
return
stan
darddeviation
insidethefamily;
Exp
.Ra
tioD
ispe
rsio
nis
thelagged
within-fam
ily
cross-sectional
stan
darddeviation
oftheexpense
ratios,
Fees
Gap
isthehighestfeeminusthelowestfeechargedwithin
thefundfamily,
fees
are
computedas
theexpense
ratioplus1/7thof
rear
andback-load
fees;Column(8)reports
coe�
cients
estimated
keepingin
oursample
only
observations
afterrules38a-1an
d206(4)-7
complian
cedate(O
ctob
er5,
2004).
Tim
ean
dStock
fixede↵
ects
areincluded
anderrors
areclustered
atthetimelevel.
Theconstan
tis
included
inallspecification
sbutthecoe�
cientis
not
reported.***,
**,*indicatestatisticalsign
ificance
atthe1%
,5%
,an
d10%
level,respectively.
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Sam
ple:
All
All
All
All
All
All
All
Post2004
WeakGovernan
ce0.0052***
0.0027***
0.0019**
(10.34)
(3.65)
(2.50)
Siblings
0.0016***
0.0015***
0.0016**
(4.88)
(3.09)
(2.25)
Fam
ilySize
0.0001
-0.0046**
-0.0063***
(0.07)
(-2.27)
(-2.88)
ReturnsDispersion
0.0584
-0.0117
-0.0304
(1.51)
(-0.42)
(-0.70)
FeesGap
0.3177***
0.1119***
0.1148***
(12.71)
(3.90)
(3.53)
FeesDispersion
1.4014***
(13.24)
Stock
Fixed
E↵ect
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Tim
eFixed
E↵ect
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Observations
738,476
738,476
738,476
738,476
738,476
738,476
738,476
309,780
R-squ
ared
0.422
0.419
0.417
0.417
0.421
0.421
0.423
0.290
50
Table
VII:Does
Cro
ss-T
radingIn
crea
seth
eDi↵eren
cein
Per
form
ance
betwee
nSta
rand
JunkFunds?
This
table
presentsresultsforregression
sof
Spre
ado
fSt
yle
Adj.
retu
rns(4-f
acto
rAl
phas)on
CT
%an
dcontrols.Eachob
servationis
obtained
from
thepairw
isecombinationsof
inflow
fundsan
dou
tflow
fundsdrawnfrom
thesamefamily,
mon
th,an
dstyle.
Thehy
potheses
tested
arenoperform
ance
shifting(b
=0),cross-subsidizationof
junkfunds(b
<0),winner-picking(b
>0).Thedep
endentvariab
leis
computedas
thereturn
(4-factoralpha)
ofinflow
fund
i(i.e.,fundswithflow
sin
thetoptercileof
family
fin
agivenmon
tht)
minustheou
tflow
fund
j’sreturn
(4-factoralpha),i.e.,funds
withflow
sin
thebottom
tercileof
family
fin
agivenmon
tht.
Fundswithflow
sin
theinterm
ediate
tercilearedropped.
CT
%is
computedas
the
percentageof
trad
esthat
arecrossedbetweensiblings
forfamily
fin
mon
tht.
Theother
indep
endentvariab
lesare:
Fam
ilySi
ze,thenaturallogof
totalassets
under
man
agem
entat
thefamilylevelin
mon
tht-1;
DSiz
e,thelogdi↵erence
betweenlagged
funds’
iand
jtotal
assets
under
man
agem
ent;
DFlo
ws,
thedi↵erence
infunds’
ian
djflow
s;DP
astF
low
s,thedi↵erence
infunds’
ian
djlagged
flow
s;DP
astR
etur
ns,thedi↵erence
infunds’
ian
dj
lagged
returns;an
dRe
turn
sD
ispe
rsio
n,themon
thly
lagged
cross-sectional
stan
darddeviation
ofreturnsinsidethefamily.
Theconstan
tisincluded
inallspecification
sbutthecoe�
cientis
not
reported.Thefrequency
oftheob
servationsis
mon
thly.Tim
e/Fam
ily/
FundPairfixede↵
ects
areincluded
when
specified
anderrors
areclustered
atthetimelevel.***,
**,*indicatestatisticalsign
ificance
atthe1%
,5%
,an
d10%
level,respectively.
Spread
ofStyle
Adj.
returns
Spread
of4-factor
Alphas
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
CT%
0.1254***
0.1431***
0.1253***
0.1536***
0.0507**
0.0619***
0.0509**
0.0678***
(3.53)
(4.11)
(3.68)
(4.47)
(2.32)
(2.79)
(2.41)
(3.11)
Fam
ilySize
-0.0012
-0.001
00.0001
-0.0004
(-1.31)
(-0.99)
(0.24)
(-0.62)
Return
Dispersion
0.0862
0.0633
0.0885**
0.0731*
(1.40)
(1.11)
(2.00)
(1.67)
DSiz
e-0.0002
-0.0027**
0.0000
-0.0014**
(-0.84)
(-2.49)
(0.30)
(-2.56)
DFlo
ws
0.0788***
0.0707***
0.0636***
0.0536***
(8.57)
(6.78)
(11.56)
(10.13)
DPas
tFlo
ws
-0.0306***
-0.0238***
-0.0254***
-0.0213***
(-3.66)
(-3.14)
(-5.07)
(-4.37)
DPas
tRet
urns
0.0071
-0.0727
-0.0084
-0.0576***
(0.11)
(-1.24)
(-0.40
)(-3.00)
FundPairFixed
E↵ect
No
Yes
No
Yes
No
Yes
No
Yes
Fam
ilyFixed
E↵ect
Yes
No
Yes
No
Yes
No
Yes
No
Tim
eFixed
E↵ect
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Observations
108,720
108,086
108,332
107,706
108,720
108,086
108,332
107,706
R-squ
ared
0.131
0.261
0.145
0.276
0.068
0.189
0.088
0.207
51
Table
VIII:
Does
Cro
ss-T
radingBoost
thePer
form
ance
ofHigh-F
eeFunds?
This
table
presentsresultsforregression
sof
Spre
ado
fSt
yle
Adj.
retu
rns(4-f
acto
rAl
phas)on
CT
%an
dcontrols.Eachob
servationis
obtained
from
thepairw
isecombinationsof
inflow
fundsan
dou
tflow
fundsdrawnfrom
thesamefamily,
mon
th,an
dstyle.
Thehy
potheses
tested
arenoperform
ance
shifting(b
=0),cross-subsidizationof
junkfunds(b
<0),winner-picking(b
>0).Thedep
endentvariab
leis
computedas
thereturn
(4-factoralpha)
ofinflow
fund
i(i.e.,fundswithflow
sin
thetoptercileof
family
fin
agivenmon
tht)
minustheou
tflow
fund
j’sreturn
(4-factoralpha),i.e.,funds
withflow
sin
thebottom
tercileof
family
fin
agivenmon
tht.
Fundswithflow
sin
theinterm
ediate
tercilearedropped.
CT
%is
computedas
the
percentageof
trad
esthat
arecrossedbetweensiblings
forfamily
fin
mon
tht.
Fees
isdefined
asE
xpen
seRa
tio+1/7(
Fro
ntLo
ad+
Rear
Load
)Theother
indep
endentvariab
lesare:
Hig
hFe
esequalson
eifafundchargesab
ovemedianfees
within
itsfamilyin
mon
thtan
dzero
otherwise;
Fam
ilySi
ze,the
naturallogof
totalassets
under
man
agem
entat
thefamilylevelin
mon
tht-1;
DSiz
e,thedi↵erence
inthenaturallogof
thelagged
funds’
iand
jtotal
assets
under
man
agem
ent;
DFlo
ws,
thedi↵erence
infunds’
ian
djflow
s;DP
astF
low
s,thedi↵erence
infunds’
ian
djlagged
flow
s;DP
astR
etur
ns,the
di↵erence
infunds’
iand
jlagged
returns;
and
Retu
rns
Dis
pers
ion,
themon
thly
lagged
cross-sectional
stan
darddeviation
ofreturnsinsidethefamily.
Theconstan
tis
included
inallspecification
sbutthecoe�
cientis
not
reported.Thefrequency
oftheob
servationsis
mon
thly.Tim
e/Fam
ily/
Fund
Pairfixede↵
ects
areincluded
when
specified
anderrors
areclustered
atthetimelevel.
***,
**,*indicatestatisticalsign
ificance
atthe1%
,5%
,an
d10%
level,respectively.
Spread
ofStyle
Adj.
returns
Spread
of4-factor
Alphas
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
CT%
xHigh-Fees
0.1769***
0.1005***
0.1460***
0.0871**
0.1247***
0.0492*
0.1042***
0.0396
(4.66)
(2.72)
(4.10)
(2.44)
(5.46)
(1.97)
(4.60)
(1.60)
CT%
0.0205
0.0829**
0.0362
0.1000**
-0.0232
0.0316
-0.0129
0.0424*
(0.45)
(2.03)
(0.81)
(2.48)
(-0.94)
(1.37)
(-0.53)
(1.80)
High-Fees
0.0029***
0.0012
0.0022**
0.0022
0.0021***
0.0012
0.0018***
0.0019
(3.64)
(0.66)
(2.41)
(1.28)
(3.39)
(0.85)
(3.26)
(1.30)
Fam
ilySize
-0.0014
-0.0009
0.0000
-0.0004
(-1.48)
(-0.96)
(0.02)
(-0.59)
Return
Dispersion
0.0889
0.0637
0.0907**
0.0736*
(1.45)
(1.11)
(2.06)
(1.68)
DSiz
e0.0001
-0.0027**
0.0003**
-0.0015**
(0.25)
(-2.51)
(2.22)
(-2.57)
DFlo
ws
0.0736***
0.0708***
0.0597***
0.0538***
(7.94)
(6.77)
(11.05)
(10.07)
DPas
tFlo
ws
-0.0297***
-0.0235***
-0.0247***
-0.0210***
(-3.55)
(-3.09)
(-4.97)
(-4.29)
DPas
tRet
urns
0.0036
-0.0728
-0.0111
-0.0577***
(0.06)
(-1.24)
(-0.52)
(-3.00)
FundPairFixed
E↵ect
No
Yes
No
Yes
No
Yes
No
Yes
Fam
ilyFixed
E↵ect
Yes
No
Yes
No
Yes
No
Yes
No
Tim
eFixed
E↵ect
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Observations
108,720
108,086
108,332
107,706
108,720
108,086
108,332
107,706
R-squ
ared
0.138
0.261
0.150
0.277
0.075
0.190
0.092
0.208
52
Table
IX:W
hatwasth
eIm
pact
ofRestrictiveReg
ulation
on
Perform
ance
Shifting?
This
table
presentsresultsforregression
sof
Spre
ado
fSt
yle
Adj.
retu
rns(4-f
acto
rAl
phas)on
CT
%an
dcontrols.Eachob
servationis
obtained
from
thepairw
isecombinationsof
inflow
fundsan
dou
tflow
fundsdrawnfrom
thesamefamily,
mon
th,an
dstyle.
Thehy
potheses
tested
arenoperform
ance
shifting(b
=0),cross-subsidizationof
junkfunds(b
>0),winner-picking(b
<0)
where
bis
thecoe�
cientof
CT
%xP
ost.
Thedep
endentvariab
leis
computedas
thereturn
(4-factoralpha)
ofinflow
fund
i(i.e.,fundswithflow
sin
thetoptercileof
family
fin
agivenmon
tht)
minustheou
tflow
fund
j’sreturn
(4-factoralpha),i.e.,fundswithflow
sin
thebottom
tercileof
family
fin
agivenmon
tht.
Fundswithflow
sin
theinterm
ediate
tercileare
dropped.
CT
%is
computedas
thepercentageof
trad
esthat
arecrossedbetweensiblings
forfamily
fin
mon
tht.
Theother
indep
endentvariab
les
are:
Post
equalson
eafterrules38a-1an
d206(4)-7
complian
cedate(O
ctob
er5,
2004),
andzero
before;
Fam
ilySi
ze,thenaturallogof
totalassets
under
man
agem
entat
thefamilylevelin
mon
tht-1;
DSiz
e,thedi↵erence
inthenaturallogof
thelagged
funds’
iand
jtotal
assets
under
man
agem
ent;
DFlo
ws,
thedi↵erence
infunds’
ian
djflow
s;DP
astF
low
s,thedi↵erence
infunds’
ian
djlagged
flow
s;DP
astR
etur
ns,thedi↵erence
infunds’
ian
dj
lagged
returns;an
dRe
turn
sD
ispe
rsio
n,themon
thly
lagged
cross-sectional
stan
darddeviation
ofreturnsinsidethefamily.
Theconstan
tisincluded
inallspecification
sbutthecoe�
cientis
not
reported.Thefrequency
oftheob
servationsis
mon
thly.Tim
e/Fam
ily/
FundPairfixede↵
ects
areincluded
when
specified
anderrors
areclustered
atthetimelevel.***,
**,*indicatestatisticalsign
ificance
atthe1%
,5%
,an
d10%
level,respectively.
Spread
ofStyle
Adj.
returns
Spread
of4-factor
Alphas
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
CT%
xPost
-0.2288***
-0.1563***
-0.1493***
-0.1713***
-0.1368***
-0.0912**
-0.0899***
-0.0996***
(-3.00)
(-2.99)
(-3.11)
(-3.61)
(-3.10)
(-2.29)
(-2.88)
(-2.81)
CT%
0.1846***
0.1545***
0.1330***
0.1662***
0.0804***
0.0686***
0.0556**
0.0752***
(3.04)
(4.21)
(3.74)
(4.60)
(2.80)
(2.95)
(2.54)
(3.31)
Post
-0.0025
-0.0019
(-1.25)
(-1.35)
Fam
ilySize
-0.0012
-0.0010
0.0002
-0.0004
(-1.30)
(-1.01)
(0.25)
(-0.64)
Return
Dispersion
0.0821
0.0587
0.0860*
0.0705
(1.34)
(1.04)
(1.96)
(1.62)
DSiz
e-0.0003
-0.0027**
0.0000
-0.0015**
(-0.85)
(-2.51)
(0.28)
(-2.58)
DFlo
ws
0.0790***
0.0709***
0.0637***
0.0537***
(8.57)
(6.79)
(11.57)
(10.13)
DPas
tFlo
ws
-0.0306***
-0.0238***
-0.0254***
-0.0212***
(-3.66)
(-3.14)
(-5.07)
(-4.37)
DPas
tRet
urns
0.0071
-0.0727
-0.0084
-0.0576***
(0.11)
(-1.24)
(-0.40)
(-3.00)
FundPairFixed
E↵ect
Yes
Yes
No
Yes
Yes
Yes
No
Yes
Fam
ilyFixed
E↵ect
No
No
Yes
No
No
No
Yes
No
Tim
eFixed
E↵ect
No
Yes
Yes
Yes
No
Yes
Yes
Yes
Observations
108,086
108,086
108,332
107,706
108,086
108,086
108,332
107,706
R-squ
ared
0.183
0.261
0.146
0.277
0.159
0.189
0.088
0.207
53
Table X: Alternative Benchmarks
This table reports the OLS estimates obtained by regressing Execution Short f all on CT Dummy and controls.
Execution Short f all in Panel A is defined as|Pj,i,t�VWAPi,d |
VWAPi,dwhere VWAPi,d is the volume weighted average price
of stock i in the day d when trade j is executed. Execution Short f all in Panel B is defined as|Pj,i,t�Openi,d |
Openi,dwhere Openi,d is the opening price of stock i in the day d when trade j is executed. CT Dummy equals oneif a trade is a cross-trade and equals zero when a trade is executed in the open market. Volume is the sharevolume of the trade; Illiquidity is Amihud’s monthly illiquidity ratio computed from daily returns obtainedfrom CRSP; Bid�Ask Spread is the di↵erence between the bid and the ask at the beginning of the month asreported from CRSP; 1/Price is 1 over the opening price of the day; Market Equity Decile is the equity decilecomputed using NYSE breakpoints; S&P500 Dummy equals one if a stock is included in the S&P500 index andzero otherwise; Stock Volatility is the within-month standard deviation of daily stock returns. Observationsare at the trade level and are obtained by drawing a 1% sample of trades from ANcerno without replacement.Stock, time, and family fixed e↵ects are included and errors are clustered at the time level. The constant isincluded in all specifications but the coe�cient is not reported. ***, **, * indicate statistical significance atthe 1%, 5%, and 10% level, respectively.
Panel A: VWAP
(1) (2) (3) (4) (5)
CT Dummy 0.0021*** 0.0021*** 0.0021*** 0.0021*** 0.0021***(11.35) (11.34) (11.34) (11.46) (11.53)
Volume -0.0002*** -0.0002*** -0.0002*** -0.0002***(-16.95) (-16.94) (-17.64) (-17.98)
Illiquidity 0.0110* -0.0025 0.0012(1.84) (-0.46) (0.28)
Bid-Ask Spread -0.0062 -0.0102(-0.60) (-1.59)
1/Price 0.0037*** 0.0025***(4.22) (3.88)
Market Equity Decile -0.0003***(-4.63)
S&P500 Dummy 0.0000(0.06)
Volatility 0.0175***(17.75)
Stock Fixed E↵ect Yes Yes Yes Yes YesFamily Fixed E↵ect Yes Yes Yes Yes YesTime Fixed E↵ect Yes Yes Yes Yes YesObservations 965,433 965,433 965,433 965,433 965,433R-squared 0.189 0.190 0.190 0.193 0.207
54
Table X Continued:
Panel B: Open Price
(1) (2) (3) (4) (5)
CT Dummy 0.0017*** 0.0016*** 0.0016*** 0.0016*** 0.0016***(4.97) (4.72) (4.72) (4.94) (4.96)
Volume 0.0003*** 0.0003*** 0.0003*** 0.0002***(13.56) (13.55) (12.20) (10.00)
Illiquidity 0.0161 -0.0386** -0.0247(0.51) (-2.54) (-1.65)
Bid-Ask Spread -0.0117 -0.0344(-0.30) (-1.32)
1/Price 0.0163*** 0.0121***(4.99) (5.14)
Market Equity Decile -0.0011***(-4.97)
S&P500 Dummy 0.0005*(1.72)
Volatility 0.0529***(14.34)
Stock Fixed E↵ect Yes Yes Yes Yes YesFamily Fixed E↵ect Yes Yes Yes Yes YesTime Fixed E↵ect Yes Yes Yes Yes YesObservations 949,254 949,254 949,254 949,254 949,254R-squared 0.201 0.201 0.201 0.208 0.224
55
Table XI: Is it just Commissions?
This table reports OLS estimates obtained by regressing Execution Short f all on CT Dummy and controls.
Execution Short f all is defined as follows: Execution Short f all j,i,t =|Pj,i,t�Pi,t |
Pi,t+%commissions, where Pj,i,t is the
execution price of trade j, in stock i, at execution time t; while Pi,t is the price of stock i in the market attime t. CT Dummy equals one if a trade is a cross-trade and equals zero when a trade is executed in the openmarket. Volume is the share volume of the trade; Illiquidity is Amihud’s monthly illiquidity ratio computedfrom daily returns obtained from CRSP; Bid �Ask Spread is the di↵erence between the bid and the ask atthe beginning of the month as reported from CRSP; 1/Price is 1 over the opening price of the day; MarketEquity Decile is the equity decile computed using NYSE breakpoints; S&P500 Dummy equals one if a stockis included in the S&P500 index and zero otherwise; Stock Volatility is the within-month standard deviationof daily stock returns. Observations are at the trade level and are obtained by drawing a 1% sample oftrades from ANcerno without replacement. Stock, time, and family fixed e↵ects are included and errors areclustered at the time level. The constant is included in all specifications but the coe�cient is not reported.***, **, * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Execution Shortfall + Commissions
(1) (2) (3) (4) (5)
CT Dummy 0.0013*** 0.0011*** 0.0011*** 0.0012*** 0.0012***(3.38) (2.84) (2.84) (3.05) (2.93)
Volume 0.0005** 0.0005** 0.0005** 0.0004**(2.11) (2.11) (2.06) (1.98)
Illiquidity 0.1535*** 0.0663* 0.0710**(3.25) (1.94) (2.06)
Bid-Ask Spread -0.0426 -0.0434(-0.37) (-0.40)
1/Price 0.0240* 0.0221*(1.89) (1.78)
Market Equity Decile -0.0003(-1.19)
S&P500 Dummy 0.0000(0.07)
Volatility 0.0337***(3.77)
Stock Fixed E↵ect Yes Yes Yes Yes YesFamily Fixed E↵ect Yes Yes Yes Yes YesTime Fixed E↵ect Yes Yes Yes Yes YesObservations 964,972 964,972 964,972 964,972 964,972R-squared 0.002 0.002 0.002 0.003 0.003
56
All volumes are available on our website www.bis.org.
Previous volumes in this series
No Title Author
576 August 2016
Crises and rescues: liquidity transmission through international banks
Claudia Buch, Cathérine Koch and Michael Koetter
575 August 2016
Housing collateral and small firm activity in Europe
Ryan N Banerjee and Kristian S Blickle
574 August 2016
Low long-term interest rates as a global phenomenon
Peter Hördahl, Jhuvesh Sobrun and Philip Turner
573 July 2016
Intraday dynamics of euro area sovereign credit risk contagion
Lubos Komarek, Kristyna Ters and Jörg Urban
572 July 2016
Housing Prices, Mortgage Interest Rates and the Rising Share of Capital Income in the United States
Gianni La Cava
571 July 2016
On the transactions costs of quantitative easing
Francis Breedon and Philip Turner
570 July 2016
Unconventional monetary policies: a re-appraisal
Claudio Borio and Anna Zabai
569 July 2016
Monetary policy, the financial cycle and ultra-low interest rates
Mikael Juselius, Claudio Borio, Piti Disyatat and Mathias Drehmann
568 June 2016
Output gaps and policy stabilisation in Latin America: the effect of commodity and capital flow cycles
Enrique Alberola, Rocio Gondo, Marco Lombardi and Diego Urbina
567 June 2016
Understanding the changing equilibrium real interest rates in Asia-Pacific
Feng Zhu
566 May 2016
Monetary facts revisited Pavel Gertler and Boris Hofmann
565 May 2016
The Collateral Trap Frédéric Boissay and Russell Cooper
564 May 2016
Moore’s Law vs. Murphy’s Law in the financial system: who’s winning?
Andrew W Lo
563 May 2016
Who supplies liquidity, how and when? Bruno Biais, Fany Declerck and Sophie Moinas
562 May 2016
Expectations and investment Nicola Gennaioli, Yueran Ma and Andrei Shleifer
561 May 2016
Mobile collateral versus immobile collateral Gary Gorton and Tyler Muir
560 May 2016
Has the pricing of stocks become more global?
Ivan Petzev, Andreas Schrimpf and Alexander F Wagner
559 April 2016
A comparative analysis of developments in central bank balance sheet composition
Christiaan Pattipeilohy